Microsoft Research: editing under 10 neurons can backdoor an AI agent
A trigger string makes the model run an attacker command with the agent's access; Microsoft has not seen real-world use.
62
- Mentions · 7d1
- Mentions · 30d1
- All mentions1
- First seenNov 10, 2026
- Last mentioned4h ago
Timeline
1 articles · newest firstAppears with
- Azure Machine Learningproduct · 1
- Microsoftvendor · 1
- Anthropicorganization · 1
- Opus 4.5ai model · 1
- Microsoft Foundryproduct · 1
- MITorganization · 1
- Microsoft Defender for Cloudproduct · 1
- Microsoft Researchorganization · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.