> TritonInfoSec News
Home / Nov 12, 2026 / Story
0
#4 The Hacker News general November 11, 2026 at 07:21 UTC

Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons

By [email protected] (The Hacker News)

AI Summary

Microsoft researchers demonstrated that an AI model can be covertly backdoored by modifying fewer than 10 rarely activated neurons — without retraining — causing the model to execute attacker-specified commands when a hidden trigger is present. The technique requires write access to model weights and was tested on production-scale models, posing a novel supply-chain threat for organizations deploying third-party AI models in agentic workflows.

Read full article → https://thehackernews.com/2026/11/microsoft-resear…

Relevance score: 87.0/100

# More from November 12

  1. 1
    Microsoft Patches 1,259 Flaws, Including Exploited Outlook Bug That May Expose NTLMv2 Hashes The Hacker News
  2. 2
    Who Runs the Ransomware Group Qilin? Krebs on Security
  3. 3
    Active ScreenConnect attacks breach MSPs, impact dozens of organizations BleepingComputer
  4. 5
    Cisco BroadWorks and Openwave Flaws Actively Exploited Despite Available Patches The Hacker News
  5. 6
    Critical Xen Orchestra auth bypass flaw exploited to hijack VM servers BleepingComputer
  6. 7
    Fortinet Patches Nearly 100 Vulnerabilities SecurityWeek
  7. 8
    Three China-Nexus Espionage Campaigns Linked Through Sparrow Malware Ecosystem The Hacker News
  8. 9
    eMudhra Revokes 8 More CAs It Previously Declared Unaffected by Key Compromise The Hacker News
  9. 10
    Railway Says Stolen Customer AI Keys Led to $740,000 in Estimated Unauthorized Usage Charges The Hacker News