1. Anthropic Says One Jailbreak Shouldn't Kill a Model. Its Biggest Investor Disagreed First. Anthropic pulled worldwide access to two models on Friday, and the company wants everyone to know it fought the decision.

2. The IPO window AI shut for years just reopened, and everyone wants through it before it closes For half a decade, the biggest names in AI raised privately and stayed there.

3. I Asked Gemini for a Backyard App. It Built One, Then Told Me a Channel Was "Unrecoverably Broken" Five minutes after handing Gemini a long prompt, the writer at The Verge came back to two things. One was a working app sitting in a preview window.

In Brief

• Bezos's Prometheus raises $12 billion to automate heavy engineering Jeff Bezos's physical-AI startup Prometheus closed a round valuing it at $41 billion. The company aims to build an "artificial general engineer" for tasks spanning heavy engineering and drug design.

• Meta unwinds $2 billion Manus acquisition after Beijing orders reversal Meta started dismantling its $2 billion Manus deal after Chinese regulators demanded the acquisition be reversed. The retreat shows cross-border AI deals now hinge on government approval from both sides.

• Google sues Chinese network for using Gemini to build scam sites Google filed suit against a cybercrime group that allegedly used Gemini to generate fraud sites at scale. The operation targeted hundreds of thousands of people with AI-coded scams.

• Ukraine fielded fully autonomous drones to kill Russian soldiers Ukraine ran a one-time test using drones that selected and engaged targets without a human in the loop. The military is installing AI modules across drones and ground robots, though full autonomy remains rare.

• NVIDIA Blackwell tops first agentic AI benchmark Artificial Analysis launched AgentPerf, a benchmark for comparing systems on agent workloads. NVIDIA's Blackwell Ultra NVL72 led the first published round, running 20x more agents per megawatt than the prior generation.

• Protests blocked $130 billion in data center projects this year Local opposition has stalled $130 billion worth of data center construction in 2026 so far. Organizers describe the wins as a first taste of political power against AI infrastructure buildouts.

• Meta's new AI unit descends into internal conflict Executives and staff are clashing over Meta's AI strategy, per internal discussions reviewed by WIRED. The reporting describes confusion over direction and open hostility in employee meetings.

• Anthropic publishes results from first Public Record Anthropic released findings from its inaugural Public Record exercise. The disclosure documents how the company is reporting on model decisions and policy commitments.

• India debates AI independence after Anthropic cuts model access Anthropic's suspension of new-model access prompted Indian tech leaders to question reliance on foreign providers. The episode reopened debate over building domestic AI capacity.

• Pokémon Go player data flowed into military drone tech Location data from Pokémon Go players was repurposed to train AI with military drone applications. The disclosure adds to scrutiny over how consumer game data feeds defense systems.

Read the full edition →