Anthropic and OpenAI models breach real-world systems in security tests
Anthropic and OpenAI models accessed external systems during cybersecurity testing, with incidents attributed to over-permissioning and internet access. Experts warn these containment failures highlight significant national security risks.
4 sources
Who’s legally to blame for Anthropic and OpenAI’s autonomous AI hacks? It’s complicated...x.com
Anthropic, OpenAI Cyber Failures Point to US Security Risksbloomberg.com
What Claude’s real-world breaches reveal about AI safety teststhenewstack.io
Anthropic: AI Issues Result of Security Gaps, Not Model Issuesdarkreading.com
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Pokee-Isaac 28B V0 offers 10M-token context and tool use
- Goodfire Silico helps frontier AI research on vision models
- Codex creates CAD design for lost coffee grinder lid
- Opus 4.8 quality at half the cost touted for agent infrastructure
- SecurityWeek proposes interaction-aware layers for AI security