Friday, September 25, 2026
The request from the Office of the National Cyber Director would end the privileged early access that Britain's AI Security Institute has had to frontier models.
Anthropic will use Akamai Cloud's infrastructure and software to handle growth in CPU workloads, and the deal can grow by up to $9 billion more.
Sanders and Casar would freeze advanced AI development until a new cabinet-level agency sets rules, and Markey wants an independent board to investigate AI agents that escape sandboxes.
OpenAI says it found the June intrusion in August during a companywide review of misbehaving agents. Transluce has also detailed earlier probes of US data sites.
ThreatDown found the operation's tooling in an unsecured Docker registry, with operational evidence going back to October 2024.
An interim report from the Forecasting Research Institute says AI reached gold-medal level at the math olympiad five years before the median expert forecast.
The animated persona lip-syncs, changes expressions and switches among 97 languages, and for now only Gemini Enterprise customers can use it.
The company says the model sets a success-rate record on the RoboLab-120 leaderboard at less than half the size of the previous best open model.