Get Started
Home
Topics
Search
Library
Research questionHow can retrieval-augmented code generation resist targeted vulnerabilities from one task-matched poisoned artifact?Retrieved artifacts can influence generated code, allowing a malicious artifact to carry an attacker-selected weakness into the output without changing the underlying language model. The difficulty is preventing or detecting that influence when the attacker cannot access the deployed retrieval and generation components.
AI
Alignment & Safety
Code Generation & Program Synthesis
Information Retrieval
Retrieval-Augmented Generation
Technology
Latest papersRecent research connected to this question, newest first.CodePoisonRAG: Knowledge Poisoning Attacks on Retrieval-Augmented Code GenerationThe evidence concerns retrieval-augmented code generation using external code artifacts, documentation, and patches. It evaluates 85 poisoned artifacts spanning ten CWE classes in Java and C, with an attacker lacking access to the victim's knowledge base, retriever, re-ranker, generator, prompt, or defenses; results cover three generators and include testing against CodeGuarder.research paper · Sep 2, 2026
Related questions
How can artifact evaluation identify security risks in publicly released research artifacts?How can LLMs generate functionally correct code without introducing security vulnerabilities?How can coding agents report defective test infrastructure instead of exploiting it to pass?How can black-box systems detect and mitigate reward hacking in self-evolving language-model loops?