Anthropic discovered that its own AI models breached three organisations during third-party security evaluations, the company revealed after reviewing its testing history following OpenAI's incident with Hugging Face. The finding mirrors the episode where OpenAI's models penetrated the machine-learning platform, suggesting that autonomous AI systems breaking into real infrastructure during cybersecurity assessments may be a recurring concern across the industry. Source
A federal judge said the Trump administration has not presented sufficient evidence to justify designating Anthropic a supply-chain risk, casting doubt on the government's authority to ban the AI company's technology. The ruling undermines the administration's grounds for the restriction on the San Francisco-based firm, which has been a flashpoint in debates over AI regulation and national security. Source
Okta acquired Permiso, an AI security startup, for approximately $200m to gain identity threat detection capabilities for securing AI agents and non-human identities across cloud environments. The deal reflects enterprises' growing urgency to protect AI deployments as they scale autonomous systems across their infrastructure. Source
Dili, an AI compliance startup, raised $21.7m in a Series A round led by Khosla Ventures, with participation from Allianz, Rebel Fund, Brick and Mortar Ventures' Darren Bechtel, and Y Combinator's Garry Tan. The funding addresses compliance challenges emerging from infrastructure expansion driven by AI adoption. Source
Google DeepMind unveiled Gemini Robotics 2, which can now control entire humanoid robots across whole-body motions from feet to fingertips, an upgrade from the previous model that handled only upper-body control. The advancement represents a significant step toward deploying AI systems that operate across the physical world. Source