2 articles
- 4d agoMicrosoft Security BlogEarliestDetecting sleeper agents hiding in AI model neurons
- 4d agoCSO OnlinePickupRewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor
← Back to the boardNews
Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor
Why it’s here: Original specialist reporting on a Microsoft-researched AI model-weight backdoor technique bears directly on AI and cybersecurity policy, including model provenance, supply-chain integrity, and federal AI security guidance.
Trace across the boardItem
4d agoCSO Online · research by Microsoft ResearchRewriting less than 10 neurons in an LLM can turn it into a sleeper backdoorMicrosoft Research reports that editing fewer than 10 neurons can turn an AI agent model into a sleeper agent that runs attacker commands on a trigger string, and recommends weight-level audits and runtime controls as defenses. CyberAI2 linked
Coverage
2 outlets · first seen 4d agoConnected
1 links- NewsDetecting sleeper agents hiding in AI model neuronsBoth involve Microsoft Research