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
Product
Azure Machine Learning
- Mentions · 7d1
- Mentions · 30d1
- All mentions2
- First seenSep 18, 2026
- Last mentioned2h ago
Timeline
2 articles · newest firstMicrosoft Patches 18 Vulnerabilities in AI, Cloud Products
Microsoft patched 18 elevation-of-privilege and information-disclosure vulnerabilities across Azure and Copilot products; all fixed server-side with no exploitation observed.
40
Related CVEs
by mentions- Double Free in Windows Secure Kernel Mode Enables Local Privilege Escalation
Appears with
- Microsoftvendor · 2
- Azure Container Registryproduct · 1
- Opus 4.5ai model · 1
- Microsoft 365 Copilotproduct · 1
- Microsoft Foundryproduct · 1
- Azure Portalproduct · 1
- PyRITproduct · 1
- Microsoft Defender for Cloudproduct · 1
- Microsoft Dataverseproduct · 1
- Microsoft Fabricproduct · 1
- Microsoft Researchorganization · 1
- Windowsproduct · 1
- Azureproduct · 1
- Microsoftorganization · 1
- Azure Cosmos DBproduct · 1
- MITorganization · 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.