AINewsnow

Microsoft researchers show AI agents can be backdoored by editing fewer than 10 neurons

This story is from 2026-11-11. It is preserved in the archive; the latest stories are on the live feed.

Microsoft researchers found that dormant, rarely activated neurons in large language models can be edited — fewer than 10 — to implant a hidden backdoor in an otherwise normal AI agent.

Read the full story at The Next Web ↗

Timeline · 1 report

  1. 2026-11-11 08:30 · The Next Web
    Microsoft finds AI agents can be backdoored by editing under 10 neurons

More stories

  1. Tech Allies Rally Behind Jensen Huang, Calling Him Vital in A.I. Race — New York Times Technology
  2. Microsoft restores services after day-long global outage affecting Outlook, Teams and Copilot — Euronews Next
  3. How Nvidia became so crucial to the AI boom — New Scientist Technology
  4. DuPont Partners with Microsoft and NobleAI on Agentic AI for R&D — The Globe and Mail
  5. Wave of Articles Argues 'Vibe Coding' Still Requires Real Software Engineering — DEV Community — AI
  6. MIT unveils AI copilot that seeks human feedback for structural design — MIT News AI
  7. Fish Audio Live 3 Claims 186 ms Full-Duplex Reactions — Dual-Track WASAPI Still Sets the Real Floor — DEV Community — AI
  8. Nvidia, Alphabet, Microsoft, and Micron Are All Buying Anthropic’s IPO. Should You Buy Too? — The Globe and Mail

Get the daily brief of stories like this at 6:30 every morning →