Microsoft researchers show AI agents can be backdoored by editing fewer than 10 neurons
Microsoft researchers showed a technique called Lazarus that plants a hidden backdoor in an AI model by changing fewer than 10 rarely used neurons, with no retraining. When a trigger appears, an agent using the model runs an attacker's command, yet it behaves normally during standard safety tests.
Companies that give AI agents access to terminals or systems need to verify where model weights come from and who can change them, because safety tests may not catch this kind of tampering.
aii.news picks and summarizes stories from trusted news sources with the help of AI. Every story links to the original reporting. How we work.
Keep reading
-
Anthropic releases Claude Fable 5.2 and Mythos 5.2, topping a multi-stage cyberattack benchmark
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, saying both improve on coding, reasoning and vision. Fable 5.2 beat GPT-6 Cyber on a benchmark that tests multi-stage cyberattacks.
Why it mattersStronger frontier models help developers do more, but their growing cyberattack skills raise the stakes for security teams.
-
Cursor confirms breach let hackers push a malicious Claude Code extension through its marketplace
Cursor confirmed a security breach that allowed attackers to distribute a malicious Claude Code extension through its marketplace. The attackers used credentials stolen from AI infrastructure provider Fireworks AI.
Why it mattersDevelopers who installed extensions from Cursor's marketplace should check their tools, because one stolen credential can spread through the AI supply chain.
-
Microsoft's AI bug hunter finds Rust standard library flaw hidden for nearly 11 years
Microsoft FORGE Lab researchers used AI to find a memory-safety bug in Rust's standard library that went unnoticed for almost 11 years, plus other flaws that can cause undefined behavior even in safe Rust code. Two findings affect stable Rust.
Why it mattersAI tools are now finding old bugs in trusted code, so developers should expect more patches, and attackers may find bugs this way too.
-
Claude Code command found saving credentials to public GitHub repositories
Researchers found that a command in Claude Code can accidentally save files containing credentials and other sensitive data to public Git repositories. Thousands of secrets have leaked this way.
Why it mattersTeams using AI coding agents should scan their repositories for exposed keys and rotate any that leaked.
-
Over 35,000 apps built with Grok's coding agent found open online, some leaking passwords
Security firm Claroc found 37,571 apps built with Grok online, and 35,578 of them could be opened without signing in, including 24 built by staff at X and SpaceXSI. About one in ten contained internal business data, and some leaked confidential information and passwords.
Why it mattersAnyone building apps with AI coding tools should check access settings before publishing, since default setups can expose private data to anyone.
