ai_security▌

← feed

NOTABLE

Microsoft Research: fewer than 10 neuron edits create sleeper-agent backdoors in AI models that pass evaluations

Microsoft Research threat research: editing fewer than 10 neurons can inject a short payload into a capable agent model so it passes evaluations yet executes an attacker's command when a trigger string appears in its context. The weighting-space attack needs only inference-level compute; Claude Code with Opus 4.5 discovered most of the threat autonomously and Microsoft has not observed threat actors weaponizing it.

released 2026-11-10 · ingested 2026-11-11T12:15Z

Sources

Primary source

Cite this

APA

AI Security Index. (2026). Microsoft Research: fewer than 10 neuron edits create sleeper-agent backdoors in AI models that pass evaluations. AI Security Index. https://www.microsoft.com/en-us/security/blog/2026/11/10/detecting-sleeper-agents-hiding-in-ai-model-neurons/

BibTeX

@misc{aisec-2026-11-11-ms-research-sleeper-agent-neuron-backdoors,
  title  = {Microsoft Research: fewer than 10 neuron edits create sleeper-agent backdoors in AI models that pass evaluations},
  year   = {2026},
  howpublished = {\url{https://www.microsoft.com/en-us/security/blog/2026/11/10/detecting-sleeper-agents-hiding-in-ai-model-neurons/}},
  note   = {AI Security Index item 2026-11-11-ms-research-sleeper-agent-neuron-backdoors}
}