GoodVibe: Security-by-Vibe for LLM-Based Code Generation

Maximilian Thang, Lichao Wu, Sasha Behrouzi, Mohamadreza Rostami, Jona te Lintelo, Stjepan Picek, Ahmad-Reza Sadeghi · 2026

Read GoodVibe: Security-by-Vibe for LLM-Based Code Generation by Maximilian Thang, Lichao Wu, Sasha Behrouzi, Mohamadreza Rostami, Jona te Lintelo, Stjepan Picek, Ahmad-Reza Sadeghi (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-2602-10778.txt.

Description

Maximilian Thang and six co-authors improve the security of AI-generated code at the neuron level in a 2026 paper. In fast, informal vibe coding workflows, speed wins and security requirements stay implicit, so models produce functionally correct but insecure code. Current fixes rely on costly full fine-tuning or coarse adaptations with limited control.

Their system, GoodVibe, rests on the insight that security-relevant thinking localizes to a small subset of neurons. Slope-based attribution identifies those neurons, enabling targeted intervention that improves security by default without retraining everything. The neuron-level approach offers clarity and control that coarse methods lack. It aims at code assistants that write safe code as readily as they write working code.

Catalog record

CREATORMaximilian Thang, Lichao Wu, Sasha Behrouzi, Mohamadreza Rostami, Jona te Lintelo, Stjepan Picek, Ahmad-Reza Sadeghi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-10778

Browse the full library · About the Tower of Babel