Anthropic caught throttling researchers (then apologized)

Engadget

Anthropic apologized and changed its approach after researchers discovered its Claude Fable 5 model was secretly downgrading requests for AI development tasks without disclosure. The company will now make these restrictions visible rather than implementing them covertly.

OpenAI is acquiring Ona to enable persistent, long-running AI agents in Codex that can operate securely in customer-controlled cloud environments over hours or days. This move supports enterprise deployment needs as organizations transition from experimenting with agents to production workflows requiring robust security and governance controls.

AWS Blog

AWS introduces a redesigned Amazon Bedrock console with project-based workflows, OpenAI and Anthropic API compatibility, and live documentation to streamline model evaluation and production deployment. The update supports latest GPT and Claude models on a new high-performance inference engine across 15 regions.

CNBC

The Dow hit a record high as investors rotated from AI and semiconductor stocks into traditional sectors, sparked by Broadcom's earnings miss and concerns about the AI rally's sustainability amid Middle East tensions.

Roblox open-sourced Cube, a foundation model family for generating 3D game assets from text, including models for controllable object generation and part-based synthesis. The release demonstrates Roblox's push toward accessible 3D AI tools for game development and creative workflows.