Anthropic said Thursday that an internal investigation uncovered three incidents in which its AI model Claude breached the systems of three organizations while conducting cybersecurity tests. …
Following the news that one of OpenAI’s models breached Hugging Face , Anthropic has now come out claiming that it has found three incidents where Claude managed to break out of its sandbox and get onto the open internet during cybersecurity evaluations. …
One of Anthropic& 039;s Claude models built and uploaded a malicious Python package to PyPI during a botched security evaluation, where it ran on 15 real systems and stole credentials from a security vendor. It was one of three incidents affecting real companies.
… The UK government-backed AI Security Institute reports that during a series of cybersecurity evaluations, Anthropic’s Claude Mythos 5 and OpenAI’s GPT-5.6 Sol both took “autonomous, unsanctioned action on the live internet,” including an instance where an agent attempted to upload malicious code to… …
… Both Anthropic and OpenAI say they have hired METR, another third-party AI evaluator, to conduct independent reviews of their respective cybersecurity incidents. …
… 6 minutes · 1,077 words · 7,173 characters 1.0× Anthropic disclosed three incidents in which Claude models accessed real systems during internal security tests. …
… Go deeper with TH Premium: AI and data centers Photonics and high-speed data movement is the next big AI bottleneck The data center cooling state of play Massive AI data center buildouts are squeezing energy supplies Ultra Ethernet: The data center interconnection of tomorrow The alleged incidents … …
Frontier Red Team Investigating three real-world incidents in our cybersecurity evaluations Jul 30, 2026 In a review of our cybersecurity evaluation transcripts, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation envi… …
… The incident has reignited concerns over how AI platforms communicate the extent of privacy and security of AI generated content.
… That prompted Anthropic to review 141,006 evaluation runs, uncovering three incidents across six runs dating back to April. …