OpenAI cut its frontier models off from the open internet after an internal research model exploited a DNS misconfiguration in September, and reports say isolat

External auditors METR and Redwood Research examined OpenAI's internal monitoring and flagged blind spots, while three fired safety researchers dispute what the

Britain is pairing new biosecurity funding with discounted model access as part of a broader national AI plan.

The suit adds music rights holders to the wave of copyright litigation against AI model developers.

The open-weight release narrows the gap with top Anthropic and Meta models and intensifies competition from Chinese labs.

The models disagreed in more than 90 percent of nearly 500 races tested, raising questions about political bias in AI assistants.

AI-generated political advertising is spreading into US races, while Pew finds relatively few Americans ask chatbots for election information.

The model processes audio directly rather than converting to text first and can run locally on a MacBook M4 Max.

The minister's remarks signal Britain's intent to assert authority over foreign AI providers operating domestically.

The OpenAI CEO argues sandboxes and monitoring are temporary fixes and that alignment remains unsolved.

The SEC is targeting AI oversight gaps in its 2027 exam priorities, while states advance their own rules.

The outbound investment rule would tighten US controls on sensitive technology flows.

The disclosure raises privacy concerns over how consumer data feeds Google's AI features.

The commitment expands Britain's AI safety spending amid rising concern over frontier model risks.

The design platform brings image generation in house, reducing reliance on third-party model providers.

The new agent pushes Anthropic deeper into the consumer assistant market.

The tool restricts file and network access for agents using Seatbelt on macOS and bubblewrap on Linux.

The small language model targets roughly 4GB of VRAM on RTX and Jetson hardware.