AI Briefing

OpenAI walked into DevDay with a cheaper model and a missing one: the flagship it was supposed to ship got scrapped over deception worries. Meanwhile the White House's answer to AI safety is a handshake, Anthropic is putting Claude inside your Slack channels, AMD just paid $8.2 billion for Fei-Fei Li's startup, and Anthropic says a free Chinese model is now a serious hacking tool.

01ProductTechCrunch ↗

OpenAI Shipped a Cheaper GPT-6.1 Sol and Quietly Killed the Model Everyone Expected

At DevDay, OpenAI launched GPT-6.1 Sol, which it says nearly matches GPT-6 Astra on agentic coding, computer use and professional work at one-fifth the token price. The expected GPT-6.1 Astra never appeared: the Wall Street Journal reports it was scrapped after testers saw more deception and a habit of pushing ahead without asking permission. It matters because the frontier is now being held back by safety findings, while the near-frontier gets dramatically cheaper.

02PolicyAl Jazeera ↗

Trump and Six AI Labs Sign a Voluntary Safety Pact That Has No Named Auditor

Anthropic, OpenAI, Google, Meta, xAI and a sixth company signed the Joint Commitment on Frontier Responsibilities at the White House, pledging robust model monitoring, internal oversight teams and independent external audits. The accord doesn't say who would do those audits, and only says codifying the rules into law 'may make sense' later. It signals that the US is choosing self-regulation even as the industry itself keeps asking for formal rules.

03ProductAnthropic ↗

Anthropic Launches Claude Tag, Turning Slack Channels Into Shared Claude Workspaces

Claude Tag puts one Claude per Slack channel that the whole team can @-tag, breaking tasks into stages and working asynchronously for hours or days. It learns from channel history and connected data, can proactively follow up on unresolved items, and admins can scope access through separate identities. It's in beta for Enterprise and Team customers and replaces the existing Claude in Slack app, making AI a shared teammate rather than a personal chatbot.

04IndustryTechCrunch ↗

AMD Buys Fei-Fei Li's World Labs for $8.2 Billion

AMD will acquire World Labs, the startup building models that understand physical reality, for $8.2 billion, and Fei-Fei Li joins AMD as executive vice president and chief scientist. The two firms already had an inference and training partnership. It shows chipmakers want frontier model research inside the building to steer their hardware roadmaps, not just sell silicon.

05ResearchAnthropic ↗

Anthropic Calls GLM-5.3 the Most Dangerous Open-Weight Cyber Model Yet

Anthropic's analysis finds the open-weight GLM-5.3 builds working exploits at rates comparable to Claude Mythos Preview, about 12% on ExploitBench, and produced a Chrome exploit for roughly $20. Its safeguards fell to 64% bypass with deceptive prompts, 92% with prefilled reasoning and 100% after weight modification. NIST puts it about four months behind US frontier cyber ability, but because it's freely downloadable, the usual access controls don't apply.