Get Started
Research questionHow can LLMs generate functionally correct code without introducing security vulnerabilities?LLM-generated code may satisfy stated behavior while still containing exploitable flaws. Generic security guidance does not always translate cleanly to a particular coding task, and coordinating multiple feedback roles can create excessive shared context.
Alignment & Safety
Code Generation & Program Synthesis
Multi-agent Systems
Latest papersRecent research connected to this question, newest first.MACGen: Toward Functionally Correct and Secure Code Generation via Multi-Agent CollaborationThe evidence concerns MACGen, a multi-agent code-generation framework involving planning, security analysis, code synthesis, and review. It was evaluated on CWEval and BaxBench using F&S@1, with reported gains relative to direct prompting.research paper · Sep 2, 2026
Related questions
How can LLMs generate correct routines when correctness depends on their joint runtime behavior?How can we generate executable tests that are both sound and discriminative for code-generating LLMs?How can LLMs edit another model’s code while preserving functionality and minimizing unnecessary changes?How can LLM sandbox security remain reliable when linguistic monitoring misrepresents internal computation?
Home
Topics
Search
Library