Evaluation Awareness Is Not One Capability: Evidence from Open Language Models
Nilesh Nayan*, Aishwarya Sampath Kumar*, Rishiraj Girmal*, Shivani Anilkumar*, Sankaran Vaidyanathan, David A. Nader Palacio, Reshmi Ghosh,, Soundararajan Srinivasan · 2026
Read Evaluation Awareness Is Not One Capability: Evidence from Open Language Models by Nilesh Nayan*, Aishwarya Sampath Kumar*, Rishiraj Girmal*, Shivani Anilkumar*, Sankaran Vaidyanathan, David A. Nader Palacio, Reshmi Ghosh,, Soundararajan Srinivasan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-23583.txt.
Description
Nayan and colleagues test whether language models notice when they are being evaluated, in a 2026 paper on AI safety measurement. Safety benchmarks assume that behavior under test predicts behavior in real deployment. That assumption breaks if a model detects evaluation cues and adjusts, making scores an optimistic ceiling, not a forecast. The team calls evaluation awareness the ability to recognize test conditions.
Eight experiments span thirty-seven open-weight models from seven families. Detection proves moderate and driven mainly by training: about two-thirds of models beat chance, and the best still trail human judges. Noticing a test does shift safety behavior, with hard refusals dropping under hypothetical framing. Yet internal probes keep reading awareness signals even when behavior collapses, and steering those signals moves downstream tasks. The axes barely correlate. Evaluation awareness is thus several loosely linked abilities, not one, so no single score can certify deployment safety.
Catalog record
| CREATOR | Nilesh Nayan*, Aishwarya Sampath Kumar*, Rishiraj Girmal*, Shivani Anilkumar*, Sankaran Vaidyanathan, David A. Nader Palacio, Reshmi Ghosh,, Soundararajan Srinivasan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.23583