Research · curated 14 Sep 2026
Smoke and Mirrors: Jailbreaking LLM-based Code Generation via Implicit Malicious Prompts | Proceedings of the 2026 IEEE/ACM 48th International Conference on Software Engineering
First reported acm.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
CodeJailbreaker demonstrates that existing instruction-following safety filters can be defeated by hiding malicious intent in covert channels, exposing a gap defenders must address in LLM-based code assistants and coding copilots.
The ICSE 2026 paper introduces CodeJailbreaker, a jailbreaking approach that bypasses LLM safety mechanisms in code generation by keeping the instruction benign while encoding malicious intent implicitly in a covert channel such as a commit message. Experiments on the RMCBench benchmark show it markedly outperforms conventional jailbreaks that state malicious intent explicitly, across three code generation tasks including against models like GPT-4.