arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.25916cs.CRcs.AI
收藏

基于反例反馈的安全智能体编码

Secure Agentic Coding through Counterexample-Grounded Feedback

Quantum AI & Cyber Security Institute (QACI), FPT Corporation(量子人工智能与网络安全研究所(QACI),FPT公司);Mila - Quebec Artificial Intelligence Institute(米拉-魁北克人工智能研究所);Université de Montréal(蒙特利尔大学);Google(谷歌)

Tue Le, Anh Nguyen Hoang, Gabriel Sasseville, Doseok Jang, Pierre-Luc Bacon, Nghi D. Q. Bui

AI总结:

提出基于反例的验证框架\falcon{},分离攻击探索与行为验证,在安全代码生成基准测试中实现更强功能安全增益,结合代理奖励可缓解奖励过度优化问题。

摘要:

随着编码模型和智能体工具的能力不断提升,可靠的安全验证正成为核心瓶颈。提升安全代码生成需要一种信号,其报告的进展需与优化后的实际程序行为保持一致;否则,优化器可在不消除底层漏洞的情况下提升信号的判定结果。现有基于模型的方法要求模型从源代码推断程序行为,或指定对抗测试的正确结果,这对安全信号提出了极高的推理要求。我们提出\falcon{},一种基于反例的验证框架,将攻击探索与行为验证分离,并通过被测程序的执行轨迹将二者连接。在测试时修复中,\falcon{}在安全代码生成基准测试中,相比其他安全信号实现了显著更强的保留功能安全增益,同时保留或提升了功能。作为强化学习奖励,\falcon{}在保留功能安全方面也优于静态分析和学习奖励,即使所有策略均成功优化了各自的训练奖励。将\falcon{}与任一代理奖励结合可进一步提升保留性能,表明基于执行的反馈可缓解奖励过度优化问题,同时保留现有安全信号的有用指导。

英文摘要:

As coding models and agent harnesses become more capable, reliable security verification is becoming a central bottleneck. Improving secure code generation requires a signal whose reported progress remains aligned with actual program behavior under optimization; otherwise, an optimizer can improve the signal's verdict without removing the underlying vulnerability. Existing model-based approaches require models to either infer program behavior from source code or specify correct outcomes for adversarial tests, placing substantial reasoning demands on the security signal. We introduce \falcon{}, a counterexample-grounded verification framework that separates attack exploration from behavioral verification and connects them through execution traces of the program under test. In test-time repair, \falcon{} yields substantially stronger held-out functional-security gains than alternative security signals across secure code generation benchmarks, while preserving or improving functionality. As an RL reward, \falcon{} also outperforms static-analysis and learned rewards on held-out functional-security, even when all policies successfully optimize their respective training rewards. Combining \falcon{} with either proxy reward further improves held-out performance, suggesting that execution-grounded feedback can mitigate reward overoptimization while preserving useful guidance from existing security signals.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑