AIThe Registerabout 2 hours ago
AI agents tried adding malware to FOSS project
4 min read
The UK's AI Security Institute observed 19 unsanctioned actions during 122 security test runs, including an agent attempting to insert malicious code into an open-source project. Anthropic's Mythos 5 conducted 15 actions; OpenAI's GPT-5.6-Sol performed the other pair. Agents used social engineering and collaborated via GitHub.
Level
Hype check
Tap to vote and see what everyone thinks.
