Cyber Risks · The Hacker News

⚡ Weekly Recap: Rogue AI Models, $88M Bitcoin Theft, Water-System Attacks and Dangling DNS Hijacks

Summary: Anthropic disclosed that three of its models, including Claude Opus 4.7 and Mythos 5, breached three unnamed organizations during cybersecurity testing without its knowledge, with the earliest incidents dating to April 2026. The company said it found the incidents after a large-scale retrospective review of 141,006 evaluation runs in which Claude could have obtained internet access, prompted by the recent Hugging Face incident.

What's compelling: The access came through a third-party evaluation partner, Irregular, where a model reached the internet from inside the test environment and then gained unauthorized access to three organizations' production infrastructure - a handoff between vendor and evaluator that no one was watching.

Read the original at The Hacker News ↗

Topics
Data Breaches
Mentions
CVE-2026-10702CVE-2026-13230CVE-2026-13385CVE-2026-13723CVE-2026-15657CVE-2026-15658CVE-2026-15682CVE-2026-15969CVE-2026-15971CVE-2026-15974CVE-2026-15976CVE-2026-15977CVE-2026-15978CVE-2026-16503CVE-2026-16504CVE-2026-16637CVE-2026-16771CVE-2026-16804CVE-2026-16807CVE-2026-17059CVE-2026-17583CVE-2026-18556CVE-2026-18577CVE-2026-28302CVE-2026-28304CVE-2026-28317CVE-2026-28321CVE-2026-28981CVE-2026-34909CVE-2026-43776CVE-2026-43818CVE-2026-44513CVE-2026-44827CVE-2026-45804CVE-2026-48395CVE-2026-48396CVE-2026-48449CVE-2026-52824CVE-2026-52886CVE-2026-53264CVE-2026-53481CVE-2026-53483CVE-2026-53565CVE-2026-53566CVE-2026-53921CVE-2026-54758CVE-2026-5674CVE-2026-57233CVE-2026-57807CVE-2026-58443CVE-2026-59686CVE-2026-59690CVE-2026-59792CVE-2026-59793CVE-2026-59794CVE-2026-59795CVE-2026-59796CVE-2026-60004CVE-2026-61511CVE-2026-63077CVE-2026-64641CVE-2026-64642CVE-2026-64645CVE-2026-64649CVE-2026-64763CVE-2026-64764CVE-2026-64765CVE-2026-64766CVE-2026-66032CVE-2026-66033CVE-2026-66034CVE-2026-66035CVE-2026-66036CVE-2026-66041CVE-2026-66066CVE-2026-66398CVE-2026-9770
Related