OpenAI said it has slowed development of Astra, an in-development AI model, after determining the system reached its "critical cybersecurity threshold"—meaning it could independently identify and carry out cyberattacks against traditionally well-protected real-world systems. The company has paused internal activities around the model pending new security standards it is putting in place. The announcement follows OpenAI's recent disclosure that its models accidentally hacked Hugging Face, a machine-learning repository. Source
Cloudflare unveiled Kitesurf, a cloud-hosted browser designed for AI agents rather than people. The tool uses less computing power than Chromium for common automation tasks, allowing developers to build browser-based AI agents more efficiently. The launch reflects growing demand from the AI agent development community for purpose-built infrastructure. Source
A New Mexico judge ordered Meta to pay $567 million in a child safety case, on top of a $375 million fine a jury had previously imposed after finding Meta misled users about the safety of its platforms. Source
Scientists used AI systems to create 16 new viruses, opening possibilities for combating bacterial resistance while raising concerns about the pace at which technology is outstripping regulation. The work demonstrates both the potential and the risks of deploying AI systems in domains where the creation of dangerous biological materials is possible. Source
Unitree Robotics, a Chinese firm, priced an IPO aiming to raise about $900 million, betting that investors are ready for humanoid robots. The offering tests market appetite for the technology, which has demonstrated impressive capabilities but has yet to prove commercial viability at scale. Source