Exposure, Extinction, and Cognitive Appraisals in Autistic Adults With Social Anxiety: Evidence From a Structured Speech Exposure Task Authors: Bruna B Roisenberg, Kelsie A Boulton, Emma E Thomas, Dorothy Yu, Adam J Guastella Published: 2026 ABSTRACT ABSTRACTSocial anxiety (SA) is highly prevalent among autistic adults, yet little is known about how autistic people respond to common therapeutic strategies, such as exposure tasks. This study examined responses to a structured speech exposure task embedded within an 8‐week modified cognitive–behavioral therapy (M‐CBT) program. Thirty‐two autistic adults with co‐occurring SA completed pre‐ and post‐task assessments of anxiety and related cognitive appraisals, including self‐focused attention, perceived performance, appearance concerns, and threat appraisals. Participants reported appropriate engagement with the task, characterized by elevated anticipatory anxiety followed by reductions in anxiety during the speech. Responses were largely consistent with CBT models of social anxiety and with reports by those in a social anxiety disorder reference group. Participants overestimated anticipatory anxiety (predicted fear) relative to experienced peak fear during the task. Higher anxiety was associated with greater self‐focused attention and stronger threat appraisals, although anxiety correlated positively with self‐rated performance. Importantly, the extinction of fear during the speech task was associated with post‐treatment response to CBT. Among participants classified as high‐extinction, 59% showed clinically significant improvement on the Liebowitz Social Anxiety Scale, Self‐Report (LSAS‐SR), compared with 25% of those in the low‐extinction group. A parallel pattern was observed in a social anxiety disorder comparison group, indicating that the extinction–outcome relationship was similar across autistic and non‐autistic adults. This study provides evidence that autistic individuals report fear and anxiety responses in a manner largely consistent with CBT models and that extinction‐based processes during exposure are meaningfully associated with treatment response. Further studies employing randomized controlled trials and objective measures of fear are now needed.Keywords: adults, anxiety, depression autistic, psychiatry, social anxiety Background Social anxiety (SA) is a significant mental health challenge for many autistic adults, contributing to marked distress and functional impairment across social, educational, and occupational domains. While approximately 12% of the neurotypical population meet diagnostic criteria for SA (Ruscio et al. 2007), prevalence estimates among autistic individuals are substantially higher. Studies report rates ranging from 29% to 57% in autistic youth (Bellini 2004; Kuusikko et al. 2008; Simonoff et al. 2008; Spain et al. 2018) and between 13% and 80% among clinically referred autistic adults (Boulton and Guastella 2021; Hofmann and Otto 2008; Joshi et al. 2012; Ketelaars et al. 2007; Lugnegård et al. 2011; Russell et al. 2015). Maddox and White (2015) estimate that nearly half of cognitively able autistic adults would meet criteria for co‐occurring SA. Despite this high prevalence, empirical evidence guiding treatment remains limited, and theoretical models developed for non‐autistic populations may not fully capture the autistic experience of social anxiety (Lei et al. 2025; Wilson and Gullon‐Scott 2024). Cognitive–behavioral models of SA, such as those proposed by Clark and Wells (1995) and Rapee and Heimberg (1997), have been foundational in understanding and treating SA. These frameworks emphasize processes, including self‐focused attention, negative self‐imagery, and threat appraisals, as key maintenance factors. They underpin cognitive–behavioral therapy (CBT), which is widely recognized as the gold‐standard treatment for SA and typically incorporates exposure‐based tasks, such as public speaking, to challenge maladaptive cognitions and reduce anxiety. However, there may also be unique challenges faced by autistic people. Fears of negative evaluation, while common, may reflect accumulated lived experiences of exclusion and misunderstanding that inform social threat appraisal (Crompton et al. 2020, 2022; Lei et al. 2025). Similarly, behaviors often labeled as safety strategies, including reduced eye contact, camouflaging, or rehearsed speech, may support adaptive communication or sensory regulation and are not necessarily indicative of anxiety‐driven avoidance (Johnson et al. 2025; Wong and Rapee 2016). Furthermore, sensory sensitivities, differences in attentional focus, and atypical patterns of interoceptive awareness may also influence anxiety and behavior (Glod et al. 2017; Kinnaird et al. 2019; Wilson and Gullon‐Scott 2024). For this reason, there has been great debate about whether the tools and measures traditionally used in the evaluation of a threat response are relevant for autistic adults. Speech tasks have long been considered a cornerstone of SA research and treatment because they reliably elicit cognitive, affective, and behavioral processes central to social anxiety, including fear of negative evaluation, self‐focused attention, and physiological arousal (Abbott and Rapee 2004; Chen et al. 2015; Guastella et al. 2009; Rapee and Abbott 2007). These tasks serve as both therapeutic exposures and experimental probes of anxiety mechanisms. A substantial body of work in non‐autistic populations demonstrates that task‐specific cognitive processes, such as attentional focus, perceived performance, and threat appraisals, predict variations in state anxiety during speech tasks (Rapee and Abbott 2007). However, little is known about how autistic adults respond to these exposures or whether the same mechanisms operate as predicted by traditional models. This gap limits theoretical knowledge and constrains the development of tailored interventions that account for autism‐specific cognitive and sensory processes. Emotional processing theory (Foa and Kozak 1986) is one of the seminal theories that explains the benefit of exposure and extinction to fear and therapeutic change. This theory conceptualizes fear memories that comprise information about feared stimuli (sensory and linked conceptual information), associated physiological and behavioral responses, and the threat meanings attached to those responses (e.g., danger, rejection, humiliation). For exposure to be effective, three conditions must be met: (1) activation of the fear structure as indicated by increased arousal, (2) extinction of the feared memory as indicated by a reduction in arousal upon exposure to the feared stimuli, and (3) incorporation of new information that disconfirms threat‐related expectations. A key indicator of successful emotional processing is within‐session fear reduction, often referred to as extinction, which reflects a decline in anxiety while the individual remains engaged in the feared situation (Baker et al. 2010; Foa and Kozak 1986). According to Foa and Kozak (1986), this reduction signals that new interoceptive and contextual information has been integrated (e.g., “anxiety can decrease without avoidance”), leading to updated threat meanings (e.g., “this situation is less dangerous than anticipated”). Within‐session extinction is therefore considered a theoretically meaningful marker of exposure engagement and therapeutic response. Despite extensive investigation of extinction processes in neurotypical populations (Abramowitz 2013; Graham and Milad 2011; Heinig et al. 2017), to our knowledge, no research to date has examined whether autistic adults show comparable patterns of within‐task anxiety reduction during social exposures, or whether extinction predicts broader improvements in social anxiety symptoms. Examining extinction during a structured speech task thus provides a critical test of whether emotional processing mechanisms operate similarly, or differently, in autistic adults. To address these gaps, the present study embedded a structured speech task within a modified CBT group program for autistic adults with co‐occurring social anxiety. Drawing on established cognitive–behavioral models, we examined whether the speech task reduced state anxiety and activated mechanisms consistent with these models. Based on prior findings in neurotypical samples (Chen et al. 2017; Rapee and Abbott 2007), we predicted that participants' state anxiety would decrease from pre‐ to post‐task. Furthermore, we expected that higher anxiety would be associated with greater self‐focused attention, poorer self‐rated performance, minimal association with appearance concerns, and stronger threat appraisals. We also anticipated that greater self‐focused attention would correlate with stronger threat appraisals and that poorer self‐rated performance would relate to stronger threat appraisals. In addition, we explored whether within‐task extinction predicted reductions in broader social anxiety symptoms across the intervention, and whether autistic adults differed from non‐autistic individuals in cognitive responses to social exposure by comparing their performance with a SA‐only sample who completed the same speech task within a comparable intervention (Rapee and Abbott 2007). Methods Participants Autism and Social Anxiety Group Thirty‐two autistic adults (age range: 18–40 years; M = 26.00, SD = 6.03) completed pre‐ and post‐intervention assessments. Recruitment occurred between March 2023 and June 2024 via clinical referrals, community sources, Headspace centers, and the Clinic for Autism and Neurodevelopment (CAN) Research at the University of Sydney. Inclusion criteria were: (a) meeting criteria for autism or autism spectrum classification on the Autism Diagnostic Observation Schedule, Second Edition (ADOS‐2) Module 4 administered as part of the research protocol. None of the participants had a prior formal autism diagnosis before entering the program. In addition to the ADOS 2, all participants met DSM‐5 criteria for Autism Spectrum Disorder based on the clinical judgment of a trained assessor following an unstructured clinical interview. The purpose of this interview was to confirm autism diagnostic status and to obtain relevant developmental and clinical background information relevant to the DSM‐5 criteria. Participants who did not meet criteria for autism based on the ADOS 2 and/or the clinical judgment of the assessor were not referred into the study. Assessments were conducted by senior researchers and clinicians with formal training in autism diagnosis and administration of the ADOS‐2, and with extensive experience working with autistic adults. Based on ADOS 2 classifications, 37.5% of participants met the assessment cut‐off for autism, and 62.5% met the cut‐off for autism spectrum. (b) A diagnosis of social anxiety disorder was confirmed using the anxiety disorders interview schedule (ADIS‐IV; Brown et al. 1994). Exclusion criteria included current suicide risk, severe depression, psychosis, substance abuse, full‐scale IQ < 70 (Wechsler Test of Adult Reading), or limited English proficiency. Demographic characteristics for this group are presented in Table 1. The mean estimated full‐scale IQ was 113.00 (SD = 6.75; range = 94–119). Gender identity distribution was men = 43.8%, women = 46.9%, other = 3.1%, and prefer not to disclose = 6.3%. Sex assigned at birth was male = 46.9%, female = 53.1%. TABLE 1Sample characteristics for autism + social anxiety and social anxiety‐only groups.CharacteristicSA + ASC (N = 32)SA‐only (N = 61)Range/%Age (years), M (SD)26.00 (6.03)27.56 (10.61)SA + ASC: 18–40; SA‐only: 15–63Estimated full‐scale IQ, M (SD)113.00 (6.75)—94–119Sex at birthMale: 46.9%; Female: 53.1%Male: 67.2%; Female: 32.8%— Note: Em dashes indicate data not available or not assessed.Abbreviations: M, mean; SD, standard deviation. Sample characteristics for autism + social anxiety and social anxiety‐only groups. Note: Em dashes indicate data not available or not assessed. Abbreviations: M, mean; SD, standard deviation. At baseline, participants completed the Liebowitz Social Anxiety Scale (LSAS; Kanai et al. 2011; Liebowitz 2015), a validated measure of social anxiety in autistic adults. The sample demonstrated high levels of social anxiety, with a mean LSAS total score of 78.75 (SD = 29.73), consistent with moderate to severe social anxiety and comparable to clinical samples. Participants also reported elevated psychological distress on the Kessler Psychological Distress Scale (K10; Kessler et al. 2003), with a mean score of 29.38 (SD = 7.60), indicating high levels of distress. Symptoms of depression, anxiety, and stress were assessed using the depression anxiety stress scales (DASS‐21; Park et al. 2019). Mean item scores suggested elevated symptomatology across domains (depression: M = 1.60, SD = 0.78; anxiety: M = 1.11, SD = 0.77; stress: M = 1.62, SD = 0.64). Baseline study details also indicated a relatively high level of educational attainment. Participants reported an average of 15.39 years of education (SD = 3.14; range = 9–21.5), with one response coded as missing. Vocational and occupational backgrounds were diverse, including participants currently enrolled in tertiary study as well as individuals engaged in part‐time and full‐time employment across professional, service, education, research, and support roles. Among those who reported weekly hours of work and/or study (n = 15), time commitment varied widely (0–48.5 h per week), with a median of 20 h per week, reflecting substantial variability in functional capacity and daily demands. A minority of participants reported receipt of unemployment benefits (6/30; 20%). Self‐reported learning and neurodevelopmental differences included attention‐deficit/hyperactivity disorder and dyslexia. Among participants who reported medication status (n = 31), approximately two‐thirds indicated current use of psychotropic medication (67.7%), most commonly stimulants (32.3%) and antidepressants (41.9%), with medication classes not mutually exclusive. General Social Anxiety Group Archival data from adults diagnosed with social anxiety disorder (SA‐only group; N = 61) were included for comparison. These data were collected as part of a university‐based treatment program conducted between 2008 and 2010, in which participants completed the same speech task and core questionnaire measures under comparable procedures. Importantly, the current modified program for autistic individuals was originally adapted from the standard CBT program used for these participants (Clarke et al. 2012). The mean age was 27.56 years (SD = 10.61; range = 15–63). Diagnoses were established using the Anxiety Disorders Interview Schedule for adults (ADIS‐IV; Brown et al. 1994). Participants were not assessed for autism as the research program at the time focused exclusively on social anxiety disorder; however, individuals who reported an existing autism diagnosis were excluded from participation. Demographic characteristics for the social anxiety‐only group are shown in Table 1. A total of 6 CBT groups for autistic adults and 11 CBT groups for social anxiety participants without reported autism were conducted at the University of Sydney Brain and Mind Center. The University of Sydney Human Research Ethics Committee approved the study (Approval No. 2015/365), and all participants provided written informed consent before taking part in the study. Intervention Program—The Engage Program The modified CBT program (Guastella and Thomas 2020; Bemmer et al. 2021) that was implemented in this study was derived from established treatment programs for adults with social anxiety (Heimberg 2002; Hofmann and Otto 2008; Rapee et al. 2009). The “Engage Program” integrated fundamental CBT elements such as exposure, cognitive restructuring, and in‐session behavioral experiments. Additionally, it encompassed the planning and evaluation of personalized homework tasks. The program used here has been previously described (see Bemmer et al. 2021). The group sessions occur over eight consecutive weeks, accommodating groups of 6–8 participants. Each session has a duration of 3 h. Participants have the option to designate a social support person who can assist with the contents being practiced during group sessions, homework tasks, and contribute to the applicability of the CBT strategies that were presented in the group session across various social situations. Five participants elected to nominate a social support person. A standard group CBT program based on established protocols (Heimberg 2002; Hofmann and Otto 2008; Rapee et al. 2009) was also delivered to participants with social anxiety who did not report an autism diagnosis. The primary procedural difference between programs concerned the timing of the speech exposure task, which was scheduled for Session 6 in the ASC groups and Session 3 in the SA‐only group. This timing adjustment allowed additional preparatory work for autistic participants, including building predictability, providing cognitive scaffolding, and introducing strategies to support sensory and communication needs, while preserving the core principles of CBT. Aside from this timing difference, the adapted and standard programs were comparable in overall structure, including the number and duration of sessions, group size, and delivery of core CBT components. Consistent with previous descriptions (Bemmer et al. 2021), the adapted program implemented cognitive restructuring in a more structured and directive manner and placed less emphasis on downward‐arrow techniques or core belief work, instead incorporating additional content related to social structures and autism. The Speech Task In the exposure‐based part of group therapy, participants were required to deliver a speech on a topic that was anxiety‐provoking, yet not overwhelmingly distressing, in front of their group members and therapists. Participants were encouraged to select topics that would challenge them to a degree that elicited anxiety without causing overwhelming discomfort. Various possible topics were suggested for use, including how social anxiety has impacted their life, what their interests are, and a topic in the media that they would like to talk about. They could select a topic that would help them complete the task and challenge their fear without feeling overwhelmed with anxiety. To measure the impact of the speech task, participants completed a series of questionnaires both before and after their speeches, assessing levels of anxiety, emotions, perceived performance, attentional focus, and self‐perceived appearance (Abbott and Rapee 2004; Chen et al. 2015; Rapee and Abbott 2007). Measures Baseline Assessments The Autism Diagnostic Observation Schedule–2 (ADOS‐2; Lord et al. 2000) is a standardized observational tool for assessing autism‐related behaviors. Module 4, designed for verbally fluent adolescents and adults, was used in this study. The Wechsler Test of Adult Reading (WTAR) (Wechsler 2001) is a neuropsychological evaluation tool used to estimate the Full‐Scale Intelligence Quotient (FSIQ). It assesses participants' ability to read aloud 50 irregular words of varying difficulty, providing scores with a mean of 100 and a standard deviation of 15. Speech Task Measures Anxiety State anxiety and fear responses associated with the speech task were assessed using Subjective Units of Distress ratings (SUDS). Participants rated anxiety on a 0–100 scale, with higher scores indicating greater distress (Guastella et al. 2009). The scale included numerical anchors and visual illustrations to support interpretation. In the context of the speech task, the terms fear and anxiety are used interchangeably to refer to subjective distress indexed by SUDS ratings. Anticipatory anxiety was indexed using pre‐task SUDS ratings of predicted fear during the speech. Peak fear refers to the highest level of fear experienced during the speech, rated retrospectively immediately after task completion. Before the speech task, participants completed four ratings: (1) current anxiety (“Please circle how anxious you feel right now”), (2) predicted average fear during the task, (3) predicted peak fear during the task, and (4) confidence in completing the task. Immediately following completion of the speech, participants completed three ratings: (1) current anxiety, (2) experienced average fear during the speech, and (3) experienced peak fear during the speech. No ratings were collected during the speech itself, ensuring that assessment did not interfere with task performance. Participants were familiar with this rating format, having used similar anxiety “thermometer” scales across earlier sessions and activities within the program. This prior exposure and scaffolding were intended to support comprehension and consistency of self‐report. Consistent with emotional processing theory (Foa and Kozak 1986), within‐session extinction was operationalized as the reduction in fear across the speech exposure and calculated as: “Extinction” = “peak experienced fear (SUDS)”–“post‐task anxiety (SUDS)”. Higher scores indicate greater habituation during the intervention, reflecting the integration of corrective information during exposure. This peak–post operationalisation is widely used in exposure research and is considered a valid behavioral index of emotional processing (Baker et al. 2010). To measure State Anxiety, we used the State Anxiety Inventory (STAI), Form Y‐1. The 20‐item state anxiety scale was administered pre‐ and post‐speech to measure anxiety “right now” on a 4‐point Likert scale, with higher scores reflecting greater anxiety (Spielberger 1989). Cognitive Appraisals Threat‐related cognitions were assessed using self‐report measures administered immediately after the speech task. The Speech Performance Questionnaire (SPQ; Rapee and Lim 1992) includes 16 items rated on a 5‐point scale, with higher scores indicating more positive self‐appraisals of performance. The scale showed internal consistency of α = 0.54 in the autistic population. Although internal consistency for the SPQ was modest in this sample, analyses involving this measure were retained to allow comparison with prior speech‐task studies. Findings involving performance self‐evaluation are therefore interpreted cautiously and emphasized in relation to effect direction rather than precision. The Probability and Consequence Questionnaire for Social Anxiety (PCQ‐SA; Shnier et al. 2022) comprises 10 items evaluating perceived likelihood and severity of negative social outcomes. The scale showed excellent internal consistency in the autistic population (α = 0.88). Attentional focus was measured using the attentional focus questionnaire (AFQ; Rapee and Abbott 2007), a 24‐item scale assessing whether attention was directed internally (toward oneself) or externally (toward the audience/environment), with internal consistency in the autistic population of α = 0.88. Appearance‐related appraisals were captured with the appearance scale (AS; Rapee and Abbott 2007), a 6‐item measure rated on a 4‐point scale (0 = “not at all attractive” to 4 = “very attractive”), with higher scores reflecting more positive perceptions of appearance. The scale showed excellent internal consistency in the autistic population (α = 0.95). Together, these measures provided an evaluation of anxiety‐related appraisals, attentional focus, and perceived performance during the speech task. Confidence and Self‐Evaluation Task‐specific confidence was assessed prior to the speech using a single‐item rating (“Please circle how confident you are of completing the task”), scored on the same 0–100 scale as SUDS ratings. Global self‐evaluation of performance was assessed immediately following the speech using the SPQ. Social Anxiety and Social Responsiveness Measures Liebowitz Social Anxiety Scale, Self‐Report (LSAS‐SR) The LSAS‐SR is a 24‐item questionnaire measuring fear and avoidance across social situations, with higher scores indicating greater social anxiety (Liebowitz 2015). For each situation, participants rate both their level of fear/anxiety and the frequency of avoidance on a 4‐point Likert scale. The LSAS‐SR has demonstrated strong internal consistency (Cronbach's α = 0.90 for fear and avoidance, α = 0.95 for total score) and solid convergent and discriminant validity in both clinical and non‐clinical populations (Fresco et al. 2001; Oakman et al. 2002). In autistic adult samples, reliability remains high, with α values exceeding 0.92 for all subscales (Kanai et al. 2011; Boulton and Guastella 2021). Social Responsiveness Scale, Second Edition, Adult Self‐Report (SRS‐2) The SRS‐2 (Constantino and Gruber 2012) is a 65‐item measure that evaluates social communication skills and autistic traits in adults. Raw scores are converted to T‐scores, with thresholds indicating typical functioning (≤ 59), mild (60–65), moderate (66–75), or severe (≥ 76) social difficulties. The instrument has excellent internal consistency in adult samples (Cronbach's α = 0.94) and is well‐validated for use in both autistic and non‐autistic populations. Those measures were administered separately as part of the pre‐ and post‐treatment assessment battery. Procedures Participants completed baseline assessments before commencing the 8‐week group CBT program. In Session 6, participants completed a structured speech exposure task designed to elicit social‐evaluative fear and activate relevant cognitive and physiological processes. Public speaking tasks were chosen because they reliably evoke social‐fear structures and are considered ecologically valid exposure exercises in SA research (Abbott and Rapee 2004; Rapee and Abbott 2007). Participants prepared for 5 min and delivered a short speech of 2–7 min on a self‐selected topic to group members and therapists. SUDS ratings were collected immediately pre‐speech, at the self‐reported peak fear point (measured by post‐task SUDS), and immediately post‐speech, allowing calculation of within‐task extinction. The STAI‐Y1 and all post‐task appraisal measures were completed immediately after the speech. The SA‐only comparison group completed the identical speech task and assessment measures within a standard CBT protocol, with the only procedural difference being the timing of the task (Session 3 vs. Session 6). This allowed comparison of cognitive and affective patterns across diagnostic groups while acknowledging that autistic participants completed the task after additional preparatory content. Statistical Analyses All analyses were two‐tailed with an alpha level of 0.05 and conducted using the Statistical Package for the Social Sciences (SPSS, version 30). Data were visually inspected for normality, and skewness and kurtosis values were examined. Normality assumptions were not violated. Pre‐ to post‐task changes in subjective distress (SUDS) and state anxiety were analyzed using mixed‐design ANOVAs, as described below. Pearson correlations examined associations among anxiety, attentional focus, performance ratings, threat appraisals, and appearance concerns. Bonferroni correction was applied for multiple comparisons (α = 0.0033). Post hoc power analyses (G*Power 3.1; Faul et al. 2009) indicated adequate power (> 0.85) for most planned correlations, except those involving appearance concerns (observed power = 0.34–0.52). Independent‐samples t‐tests were used only for exploratory comparisons of post‐task variables assessed at a single time point, where no baseline measure was available (e.g., threat appraisals, speech performance ratings, attentional focus). Outcomes assessed pre‐ and post‐task were analyzed using mixed‐design ANOVAs, and treatment outcomes were analyzed using mixed ANOVA and regression models that controlled for baseline symptom severity. Within‐session extinction was analyzed primarily as a continuous variable. Continuous extinction scores were used in correlational and hierarchical regression analyses examining associations with task‐related cognitive variables and change in social anxiety symptoms (LSAS change), consistent with recommendations to retain continuous information and maximize statistical power. For descriptive and clinical interpretability purposes only, extinction was additionally categorized into high‐ and low‐extinction groups in exploratory analyses examining rates of clinically significant improvement. These categorical analyses were treated as secondary and illustrative and were not used to support primary inferential conclusions. To evaluate whether extinction predicted broader symptom improvement over and above baseline severity, hierarchical linear regression models predicted LSAS change from demographic covariates and baseline LSAS severity, with extinction entered as the hypothesized process marker. Extended categorical and subgroup analyses are reported in the Supporting Information. In the hierarchical multiple regression analysis predicting LSAS change, predictors were entered in theoretically motivated blocks consistent with our prior baseline analytic approach (Bemmer et al. 2021; Roisenberg, Boulton, Thomas, and Guastella 2026): demographic covariates (age, sex) were entered first, baseline symptom severity (LSAS pre‐treatment) was entered second to account for initial severity and regression‐to‐the‐mean effects, and extinction was entered third as the hypothesized process marker. This sequencing allowed a direct test of extinction's incremental contribution while limiting the number of predictors given the modest sample size. Clinically significant improvement was operationalized as a reduction of at least one standard deviation on the Liebowitz Social Anxiety Scale (LSAS). In the present sample, this corresponded to a decrease of 27.24 points. For comparison, the LSAS‐SR clinical improvement threshold proposed by Von Glischinski et al. (2018), defined as a 28% reduction, would correspond to a 21.63‐point decrease. Chi‐square analyses were conducted to examine the association between extinction group (high vs. low) and clinically significant improvement within each diagnostic group. All statistical tests were two‐tailed with α = 0.05. Adjustments for multiple comparisons were applied when relevant. Community Involvement This research was developed through a collaborative approach with autistic community members and neurodivergent researchers, to center lived experience and community‐identified priorities. A Community Consultation Committee formed of autistic adults, including individuals with prior experience in the CBT group programme, contributed across multiple phases of the project. Committee members provided input on the study objectives, informed decisions about the structure of the intervention, and offered guidance on accessibility and acceptability. Alongside this, neurodivergent researchers were integral members of the research team, contributing as co‐researchers and co‐authors, contributing to data analysis and reviewing manuscript drafts. Through this sustained partnership, lived experience perspectives were meaningfully incorporated throughout the research process, from conception to dissemination. Results Pre‐Post Differences in Anxiety and Fear (ASC + SA Group) Autistic participants with co‐occurring social anxiety (ASC + SA group) reported high anxiety before and during the speech task, and there was evidence that their anxiety reduced by the end of the speech task, t(31) = 4.21, p < 0.001, d = 0.75. Peak fear also showed a significant reduction (M = 70.00, SD = 25.00), t(31) = 3.32, p = 0.002, d = 0.59. State anxiety measured by the STAI decreased from M = 24.90 (SD = 7.54) pre‐task to M = 19.70 (SD = 7.44) post‐task, t(30) = 4.27, p < 0.001, d = 0.69, indicating a moderate reduction in situational anxiety for the ASC + SA group. Between‐Group LSAS Comparisons Pre‐ to post‐treatment LSAS scores were analyzed using a 2 (time: pre, post) × 2 (Group: ASC + SA, SA‐only) mixed ANOVA. There was a significant main effect of time, F(1, 76) = 4.01, p = 0.049, partial η 2 = 0.050, indicating a modest reduction in LSAS scores across participants from pre‐ to post‐treatment. The time × group interaction was not significant, F(1, 76) = 0.04, p = 0.848, partial η 2 < 0.001, providing no evidence that symptom change differed by diagnostic group. The between‐subjects' main effect of Group was also not significant, F(1, 76) = 0.33, p = 0.570. To assess comparability, demographic characteristics were examined. The groups did not differ significantly in age, t(90.35) = −0.92, p = 0.36, and sex distribution did not differ significantly, χ 2(1, N = 93) = 3.62, p = 0.057. State–Trait Anxiety (STAI) Changes in STAI from pre‐ to post‐speech were examined using a 2 (time: pre, post) × 2 (group: ASC + SA, SA‐only) mixed ANOVA. There was a significant main effect of time, F(1, 90) = 23.27, p < 0.001, partial η 2 = 0.205, indicating that participants overall reported lower anxiety ratings following the speech compared to before the task. The time × group interaction was not significant, F(1, 90) = 0.04, p = 0.838, partial η 2 < 0.001, indicating that autistic adults with social anxiety and SA‐only participants showed comparable pre‐to‐post changes in self‐evaluation anxiety. The between‐subjects main effect of Group was also not significant, F(1, 90) = 0.15, p = 0.699, suggesting no overall group differences when averaged across time. Subjective Distress (SUDS) A parallel mixed ANOVA was conducted for subjective units of distress (SUDS) ratings collected immediately before and after the speech task. There was a large main effect of Time, F(1, 91) = 50.43, p < 0.001, partial η 2 = 0.357, indicating substantial reductions in subjective distress following the speech across participants. The time × group interaction was not significant, F(1, 91) = 1.83, p = 0.180, partial η 2 = 0.020, indicating that the magnitude of distress reduction did not differ between the ASC + SA and SA‐only groups. A small between‐subjects main effect of Group was observed, F(1, 91) = 6.95, p = 0.008, partial η 2 = 0.008, reflecting minor overall differences in mean distress levels across groups that were not related to differential change over time. To examine the full exposure trajectory, SUDS ratings collected pre‐speech, at peak fear, and post‐speech were analyzed using a 3‐level repeated‐measures ANOVA with Group as a between‐subjects factor. There was a significant main effect of time, F(1.08, 94.16) = 7.41, p = 0.007, partial η 2 = 0.078, indicating systematic changes in distress across the exposure, consistent with anticipatory anxiety, peak fear, and subsequent reduction. The time × group interaction was not significant, F(1.08, 94.16) = 0.83, p = 0.44, partial η 2 = 0.009, suggesting that autistic and non‐autistic participants exhibited a similar exposure‐related distress trajectory. Between‐Group Differences in Post‐Task Cognitive Appraisals Between‐group comparisons of post‐task cognitive appraisals were conducted using independent‐samples t‐tests, with Welch's correction applied where appropriate. Autistic adults with social anxiety reported significantly less favorable post‐task speech performance ratings than participants with social anxiety‐only, t(58.65) = −3.13, p = 0.003, with the ASC + SA group reporting lower scores (M = 27.31, SD = 11.83) than the SA‐only group (M = 35.20, SD = 10.46). The ASC + SA group also reported significantly lower overall threat appraisals on the Probability and Consequence Questionnaire compared with the SA‐only group, t(54.98) = −2.62, p = 0.011 (ASC + SA: M = 16.97, SD = 9.21; SA‐only: M = 22.12, SD = 8.04). In contrast, no significant between‐group differences were observed for appearance‐related concerns, t(53.93) = −0.97, p = 0.334, or attentional focus during the task, t(64.64) = −1.49, p = 0.140. Correlational Analyses (ASC + SA Group) Correlational analyses were conducted within the ASC + SA group to examine whether relationships among anxiety, attentional focus, performance, and threat appraisals aligned with traditional social anxiety models. Results are summarized in Table 2. TABLE 2Correlations among task‐based anxiety, cognitive appraisals, attentional focus, and performance self‐evaluation in the autistic social anxiety group (N = 32).Construct pair r p Task‐based anxiety (STAI‐Y)–attentional focus (AFQ)0.585< 0.001Task‐based anxiety (STAI‐Y)–performance self‐evaluation (SPQ)0.704< 0.001Task‐based anxiety (STAI‐Y)–appearance concerns (AS)−0.383nsTask‐based anxiety (STAI‐Y)–threat appraisals (PCQ‐SA)0.619< 0.001Attentional focus (AFQ)–threat appraisals (PCQ‐SA)0.5620.001Performance self‐evaluation (SPQ)–threat appraisals (PCQ‐SA)0.675< 0.001Threat appraisals (PCQ‐SA)–confidence in completing the task (SUDS 4)−0.636< 0.001 Note: Values are Pearson's r. Task‐based anxiety was assessed using the State–Trait Anxiety Inventory, State version (STAI‐Y1). Performance self‐evaluation was assessed using the speech performance questionnaire (SPQ). Confidence in completing the task (SUDS 4) reflects pre‐task confidence ratings. Significance values reflect Bonferroni‐adjusted thresholds.Abbreviations: AFQ, attentional focus questionnaire; AS, appearance scale; PCQ‐SA, probability and consequence questionnaire for social anxiety; SUDS, subjective units of distress scale. Correlations among task‐based anxiety, cognitive appraisals, attentional focus, and performance self‐evaluation in the autistic social anxiety group (N = 32). Note: Values are Pearson's r. Task‐based anxiety was assessed using the State–Trait Anxiety Inventory, State version (STAI‐Y1). Performance self‐evaluation was assessed using the speech performance questionnaire (SPQ). Confidence in completing the task (SUDS 4) reflects pre‐task confidence ratings. Significance values reflect Bonferroni‐adjusted thresholds. Abbreviations: AFQ, attentional focus questionnaire; AS, appearance scale; PCQ‐SA, probability and consequence questionnaire for social anxiety; SUDS, subjective units of distress scale. Consistent with predictions, higher state anxiety was strongly associated with greater self‐focused attention. Anxiety was also positively correlated with threat appraisals, indicating that participants who felt more anxious were more likely to overestimate the likelihood and severity of negative social outcomes. Similarly, greater self‐focused attention was linked to stronger threat appraisals. Unexpected patterns also emerged. Anxiety and threat appraisals correlated positively with self‐rated performance, contrary to the predicted negative association. These relationships are illustrated in Figure 1, which maps the observed empirical associations onto the Rapee and Heimberg (1997) cognitive model, highlighting both supported pathways and areas where findings differed from theoretical expectations. Empirical mapping of speech task findings onto the Rapee and Heimberg (1997) model of social anxiety. Red arrows indicate associations that diverged from theoretical predictions. Task‐based anxiety was indexed using the State–Trait Anxiety Inventory (STAI‐Y1). Cognitive and attentional processes included self‐focused attention (AFQ), threat appraisals (PCQ‐SA), appearance concerns (AS), and performance self‐evaluation (SPQ). The figure highlights that while core anxiety–attention–threat pathways were supported, the direction of the association between anxiety and performance self‐evaluation diverged from traditional model predictions. See Table 2 for corresponding statistical results. Extinction as a Predictor of Symptom Improvement We then tested whether within‐session extinction during the speech task was associated with symptom improvement post‐treatment. Greater within‐session extinction, treated as a continuous variable, was associated with greater improvement in LSAS scores (see regression results). For descriptive and clinical interpretability purposes, participants were also categorized into high‐ and low‐extinction groups to illustrate differences in rates of clinically significant improvement. Extinction showed a moderate‐to‐large correlation with LSAS change (r = 0.52, p = 0.004). A hierarchical multiple regression examined whether within‐session extinction predicted improvement in social anxiety (LSAS change). Age and sex were entered at Step 1, baseline LSAS severity was entered at Step 2, and extinction was entered at Step 3. Step 1 was not significant, F(2, 26) = 1.95, p = 0.163, R 2 = 0.130. Adding baseline LSAS severity did not improve model fit, ΔR 2 = 0.009, ΔF(1, 25) = 0.27, p = 0.605. In Step 3, extinction significantly improved the model, ΔR 2 = 0.203, ΔF(1, 24) = 7.40, p = 0.012. In the final model, extinction was a significant predictor of LSAS change, B = 0.513, SE = 0.188, β = 0.472, t(24) = 2.72, p = 0.012, 95% CI [0.124, 0.902]. The overall model was significant, F(4, 24) = 3.12, p = 0.033, R 2 = 0.342 (adjusted R 2 = 0.233). Analyses were based on N = 29 due to listwise deletion. Regression assumptions were evaluated. Multicollinearity was not evident (tolerance = 0.829–0.923; VIF = 1.084–1.207; condition indices ≤ 17.09). Standardized residuals were within acceptable limits (−1.60 to 2.16), and inspection of residual plots suggested no major departures from linearity or homoscedasticity. Given the modest sample size, the analysis is primarily sensitive to medium‐to‐large effects. In this context, the observed incremental contribution of extinction falls within the medium‐to‐large range, indicating that the model had adequate sensitivity to detect an extinction effect of this magnitude, whereas smaller incremental effects for covariates may not have been detectable. A post hoc sensitivity analysis based on the observed effect size suggests that power was approximately adequate (≈0.80) to detect an effect of this magnitude at α = 0.05. Accordingly, regression findings are interpreted conservatively, with emphasis on effect sizes, confidence intervals, and clinical relevance, and should be replicated in larger samples. Exploratory Categorical Illustration of Clinical Improvement To aid clinical interpretability of the primary continuous findings, extinction was additionally categorized into high and low groups for descriptive analyses of clinically significant improvement on the LSAS‐SR (±1 SD). The association between the extinction group and clinically significant improvement did not reach conventional statistical significance, χ 2(1) = 3.25, p = 0.071; however, the pattern was clinically suggestive. Specifically, 59% of high‐extinction participants achieved clinically significant improvement compared with 25% of low‐extinction participants (odds ratio = 4.29, 95% CI [0.84, 21.76]). These categorical findings are presented for illustrative purposes only and converged in direction and magnitude with the primary regression analyses, which modeled extinction as a continuous predictor of LSAS change. Accordingly, inferential conclusions regarding the extinction–outcome relationship are based on the continuous regression models rather than the categorical comparisons. In exploratory analyses, higher extinction was also associated with more adaptive post‐task cognitive and attentional patterns, including reduced threat‐focused attention and lower harm appraisals. Given the number of related outcomes and the modest sample size, these comparisons are interpreted cautiously and are reported in full in the Supporting Information. The pattern of effects was consistent with the continuous regression findings, suggesting that greater within‐session extinction was associated with more adaptive cognitive responding following the task. Discussion This study examined responses to a structured speech task embedded within a modified CBT program for autistic adults with social anxiety disorder. We further sought to understand whether the response to these tasks was similar to that of individuals without an autism diagnosis who also met criteria for social anxiety disorder. Results showed that the speech task led to significant reductions in anxiety from pre‐ to post‐task, and that participants showed expected levels of threat cognitions and attentional focus consistent with an anxiety response. While results were comparable with a SA reference group, there was some evidence for lower threat cognitions in the autistic SA condition. As for the correlation analysis, higher state anxiety was strongly associated with greater self‐focused attention; anxiety was also positively correlated with threat appraisals, and greater self‐focused attention was linked to stronger threat appraisals. These relationships were consistent with cognitive‐behavioral models of social anxiety (Clark and Wells 1995; Rapee and Heimberg 1997). Analyses of within‐session extinction suggested that reductions in fear during the speech task were meaningfully related to broader symptom change, predicting a significant degree of symptom improvement from pre‐ to post‐therapy. This study, therefore, provides evidence of the utility of an exposure‐based speech task in leading to predictable anxiety responses and potential for extinction as a useful marker for therapeutic gain in autistic individuals with co‐occurring social anxiety. Anxiety Reduction In line with previous research, participants demonstrated a significant reduction in state anxiety following the speech task. This pattern reflects a therapeutic effect consistent with exposure principles, where repeated contact with feared situations reduces anxiety through habituation (Chen et al. 2015, 2017; De Jong et al. 2018; Landry et al. 2022; Ramnerö 2012). Participants also consistently overestimated anticipated distress before the speech task, aligning with cognitive‐behavioral models that show the role of anticipatory threat appraisal in maintaining anxiety (Black et al. 2022; Clark and Wells 1995; Rapee and Heimberg 1997; Vassilopoulos 2005). Cognitive and attentional factors shaped participants' experiences. AFQ scores indicated moderate self‐focused attention, and greater self‐focused attention was associated with stronger threat appraisals, reinforcing the role of internal attentional bias in amplifying perceived social threat. Confidence ratings were also relevant, as greater confidence in completing the task before the speech task was associated with fewer negative threat appraisal expectations, suggesting that strengthening self‐efficacy may help reduce threat sensitivity in autistic adults, as would be predicted by cognitive‐behavioral models of anxiety (Connor et al. 2019; Kumazaki et al. 2019; Ward and Esposito 2018). Group Differences in Cognitive Responses to Social Exposure The present findings largely aligned with cognitive models of social anxiety and prior exposure‐based research. Participants across groups demonstrated reductions in fear after the speech task. However, autistic participants reported relatively less favourable self‐evaluations of performance compared to those without a diagnosis of autism. In line with Rapee and Abbott's (2007) model, participants continued to endorse self‐focused attention and evaluative concerns, indicating that anxiety‐relevant cognitive processes persist during exposure. However, autistic participants reported significantly lower estimates of both the likelihood and consequences of negative social outcomes, as well as less favorable post‐task performance appraisals, compared to the SA–only group. Notably, probability and consequence questionnaire scores in the autistic group were lower than those reported in the SA‐only group. Results from our SA‐only group were consistent with Rapee and Abbott's (2007) clinical sample, suggesting attenuated perceptions of social threat despite comparable levels of state anxiety. This pattern may help explain the absence of heightened post‐task rumination or escalating distress, indicating that reductions in perceived threat may represent a particularly important mechanism of change for autistic adults undergoing exposure‐based interventions. Several factors may account for these group differences. The modified CBT program explicitly enhanced predictability, transparency, and cognitive scaffolding, consistent with neurodiversity‐informed adaptations (Bemmer et al. 2021; Roisenberg et al. 2025). By the time participants completed the speech task (week 6 of the program), they had engaged in multiple exposures, cognitive restructuring exercises, and group reflections, likely facilitating more calibrated threat expectations. In contrast, the reference sample may have had fewer opportunities for systematic threat updating. Alternatively, autistic people may experience social contexts as pervasively threatening and engage in highly safety‐focused, biologically driven adaptations to cope with this (Lawson 2020). Experimental studies show that autistic adults demonstrate good threat detection in terms of preserved perceptual threat processing (Krysko and Rutherford 2008), but more research in this field is needed. Mechanisms of Social Anxiety in Autism Correlational analyses supported key elements of established models of social anxiety (Clark and Wells 1995; Rapee and Heimberg 1997). For example, higher state anxiety was linked to greater self‐focused attention and stronger threat appraisals (Lei and Russell 2020; Spain et al. 2018). These findings suggest that autistic adults engage in heightened self‐monitoring during socially demanding tasks (Wood et al. 2021; Wilson and Gullon‐Scott 2023), potentially interacting with sensory sensitivities to amplify perceived threat (Glod et al. 2017; White et al. 2015). However, key patterns diverged from traditional expectations. Higher anxiety was associated with more positive self‐rated performance during the speech task, and better performance ratings were linked to stronger threat appraisals. Appearance concerns were weakly associated with anxiety at an uncorrected level but did not survive correction for multiple comparisons. In contrast, appearance concerns were strongly associated with poorer self‐rated performance, suggesting that appearance may influence confidence rather than anxiety per se. Interpretation of these findings should also consider the relatively weak internal consistency of the speech performance questionnaire (SPQ) in this autistic sample. Lower reliability may reflect variability in how autistic adults interpret and evaluate their own social performance, which may contribute to the observed divergence between anxiety, threat appraisal, and performance ratings. Findings suggest an opportunity for future studies to understand differences in how autistic people report their own perceptions of speech performance (Johnson et al. 2025). Even when they perceive themselves as performing well, this success may still be interpreted through the possibility of rejection, consistent with past experiences of exclusion (Bellini 2006; Crompton et al. 2020, 2022; Lei et al. 2023; Rosen et al. 2024). The weak role of appearance concerns further distinguishes autistic social anxiety from non‐autistic presentations, where physical self‐consciousness is often central (Wong and Rapee 2016). Instead, anxiety may be more closely tied to communication vulnerabilities, sensory sensitivities, and prior history of exclusion (Crompton et al. 2020; Rifai et al. 2022). Recent evidence suggests that autistic adults often generate social imagery from a “field” perspective, with distress tied to internal bodily and sensory signals rather than fears about appearance (Lei et al. 2025). This pattern may reflect why higher anxiety co‐occurs with better performance ratings, reflecting coping strategies aimed at managing internal discomfort rather than correcting perceived flaws. Extinction and Emotional Processing Theory The extinction findings offer insight into how emotional‐processing mechanisms may function during social exposure for autistic adults. Emotional processing theory (Foa and Kozak 1986) proposes that reductions in fear during exposure reflect the integration of information that contradicts threat‐based expectations, and this assumption is a cornerstone of cognitive‐behavior therapy practices. In this study, extinction showed modest associations with improvements in social anxiety symptoms, even when considered alongside other task‐related variables. Although constrained by the small sample, this pattern is broadly consistent with the idea that within‐session fear reduction may indicate partial modification of the meanings attached to social‐evaluative situations. Participants with higher extinction also tended to show more adaptive cognitive patterns (e.g., lower threat appraisals and reduced threat‐focused attention), suggesting possible shifts in perceived probability and cost of negative evaluation. Importantly, the exploratory finding that extinction predicted therapeutic improvement in autistic adults in a manner closely aligned with the SA‐only comparison group has meaningful clinical implications. Evidence that extinction processes operate similarly across groups supports the use of exposure‐based strategies in autism and reinforces the relevance of established CBT mechanisms for guiding treatment. These results point to extinction as a potential process marker worth examining in future work with larger samples and repeated exposures. This is particularly relevant given ongoing debates about whether autistic social fears are maintained by the same mechanisms assumed in traditional social anxiety models (Wilson and Gullon‐Scott 2023, 2024). Findings from behavioral research with autistic children show that exposure alone rarely changes behavior, whereas exposure paired with extinction reliably increases behavioral flexibility and tolerance to novel or aversive stimuli (Fisher et al. 2019). To our knowledge, this is the first study to explore whether autistic adults show similar patterns of within‐task anxiety reduction during social exposures, or whether extinction predicts broader improvements in social anxiety symptoms. Why CBT Models Need Adaptation for Autism Taken together, the current findings suggest several targeted refinements to CBT models that better align with the cognitive patterns observed in autistic adults. Although core mechanisms outlined in traditional social anxiety models, such as self‐focused attention, anticipatory anxiety, and threat appraisals, were clearly active, autistic participants showed lower threat appraisals but less favorable self‑evaluations of performance than non‐autistic counterparts, despite experiencing comparable levels of state anxiety. This pattern indicates that autistic adults may enter social‐evaluative situations with less catastrophizing expectations, meaning that therapeutic efforts may benefit from emphasizing predictability, confidence‐building, and clarity around coping capacity. Additionally, the atypical association between higher anxiety and higher performance ratings suggests that some autistic adults may engage in compensatory strategies that maintain functioning even during heightened distress. Clinically, this points to the value of helping individuals differentiate between internal discomfort and actual performance deficits. This study suggests that social anxiety for autistic populations may be shaped by internal sensory discomfort, uncertainty about coping capacity, heightened self‐focused attention, and histories of misunderstanding or exclusion (Crompton et al. 2020; Lei et al. 2025; Rifai et al. 2022). Finally, the finding that extinction during the task predicted symptom change reinforces the importance of structured, scaffolded exposures that allow autistic adults to experience reductions in anxiety while remaining engaged (Bemmer et al. 2021; Roisenberg et al. 2025). Clinical Application and Relation to Behavioral Experiments From a clinical perspective, the structured speech task used in the present study overlaps with behavioral experiments commonly employed in cognitive‐behavioral models of social anxiety, including the Clark and Wells (1995) framework of standardized social interaction tasks designed to elicit and test key maintaining processes. For example, Stopa and Clark (1993) used a brief videotaped conversation with a trained confederate (“stooge”) to examine the effects of self‐focused attention, safety behaviors, and negative self‐appraisals during social interaction. Within this tradition, structured social tasks are valued not only as exposure exercises, but also as behavioral experiments that provide opportunities to examine social threat predictions, differentiate internal distress from observable social outcomes, and support individuals to make sense of their anxiety, considering prior experiences of misunderstanding, rejection, or exclusion. The speech task used in this study also offered a structured opportunity to notice how anxiety unfolds in real time, to distinguish internal distress from observable social outcomes, and to explore the role of attentional focus and coping strategies during social performance. This framing is consistent with qualitative findings from the same intervention, in which autistic participants described speech‐based and role‐play tasks as most helpful when they were clearly contextualized, collaboratively supported, and followed by debriefing that focused on coping capacity and interpersonal impact rather than perceived performance deficits (Roisenberg, Boulton, Thomas, Perry, et al. 2026). When delivered in this way, structured social tasks may be particularly appropriate for autistic adults, given that social anxiety is often shaped by uncertainty, sensory demands, and prior experiences of exclusion. Emphasizing transparency, preparation, and post‐task reflection may therefore facilitate engagement and learning without requiring participants to invalidate social concerns that are grounded in lived experience. For clinicians, these findings suggest that structured speech‐based tasks can be incorporated flexibly, either as exposure exercises or as behavioral experiments, with attention to individual goals, preferences, and meanings. Clinicians may emphasize collaborative planning, selective exploration of coping strategies (including safety behaviors that serve adaptive communication or sensory‐regulatory functions), and post‐task reflection that distinguishes internal distress from observable social impact. When used in this way, structured social tasks may support engagement and learning while respecting that autistic social anxiety is often grounded in lived experience rather than distorted belief. This flexible, autonomy‐supportive approach may help ensure that such tasks are both tolerable and therapeutically useful for autistic adults. Limitations Several limitations should be acknowledged. First, the relatively small sample size (N = 32) limits statistical power and the generalizability of the findings. Although Bonferroni adjustments were applied to reduce Type I error, the study remains exploratory in nature, and results should be interpreted with caution. That said, observed effect sizes were in the moderate‐to‐large range, suggesting that the patterns identified are meaningful and warrant replication in larger and more diverse samples. The modest sample size also constrained sensitivity to small effects, particularly in subgroup and categorical analyses. While the analytic strategy was revised to prioritize parsimonious, well‐powered models, post hoc sensitivity analyses indicated adequate power to detect only medium‐to‐large effects. Accordingly, exploratory findings, especially those based on categorical comparisons, should be interpreted cautiously. The internal consistency of the SPQ was relatively low in this sample, which may reflect variability in how autistic adults interpret self‐performance and highlights the need for adapted measures. Second, the age range of the sample was relatively narrow, with most participants in their mid‐20s, restricting applicability to older autistic adults. Third, the study relied exclusively on self‐report measures. Future work should combine self‐report with objective indicators (e.g., physiological responses, behavioral ratings) and develop or adapt assessment tools that more accurately reflect autistic experiences. The use of objective indicators may be particularly helpful to understand the role of fear extinction, using multi‐method measurement of fear reduction (e.g., psychophysiology, behavioral markers), and repeated exposures across sessions may help clarify whether extinction functions as a reliable process marker for autistic adults engaging in exposure‐based interventions. Third, participants in the social anxiety (SA‐only) group were not assessed for autism, as the original research focused exclusively on social anxiety disorder; however, individuals who reported an autism diagnosis were excluded. This means the presence of undiagnosed or unreported autistic traits cannot be entirely ruled out. While it is unlikely that undiagnosed autism was a significant feature of the SA‐only group, it cannot be ruled out in some participants. Another important limitation relates to the timing and structure of the speech task within the intervention. For autistic participants, the CBT program was adapted to include additional preparatory content aimed at supporting engagement and addressing autism‐specific needs. As part of these modifications, the speech exposure task was scheduled later in the program (Session 6) rather than earlier (Session 3), as in the standard CBT protocol for the SA‐only group. This adjustment allowed for greater predictability, cognitive scaffolding, and skill‐building before exposure, but it also introduces a confounding factor. Differences in timing and preparatory content make it difficult to determine whether observed outcomes reflect the speech task itself or broader therapeutic gains achieved through the modified program. This difference may also have contributed to lower negative appraisals of speech performance. Future research should consider implementing the speech task both before and after treatment, using consistent timing across groups, to better isolate its independent effects. Conclusion The present study provides preliminary evidence that structured exposure tasks, such as public speaking, can be integrated within CBT to support autistic adults with social anxiety. The findings also suggest that while established mechanisms of social anxiety, such as self‐focused attention and threat appraisals, remain relevant, autistic experiences are additionally shaped by compensatory effort and histories of exclusion. Incorporating these autism‐specific processes is essential for developing neurodiversity‐informed models of social anxiety that better capture the interplay between attention, performance, and perceived threat. Future research should evaluate the durability of these effects and refine theoretical models through longitudinal designs, alongside developing autism‐sensitive assessment tools to ensure outcomes are measured meaningfully. Together, these steps will strengthen the integration of neurodivergent‐affirming approaches into evidence‐based CBT, advancing interventions that are both effective and responsive to the lived experiences of autistic adults. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Data S1: aur70268‐sup‐0001‐Supinfo.docx. Acknowledgments Open access publishing facilitated by The University of Sydney, as part of the Wiley ‐ The University of Sydney agreement via the Council of Australasian University Librarians. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. References Associated Data Supplementary Materials Data S1: aur70268‐sup‐0001‐Supinfo.docx. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.