1. You're now the agent's supervisor, not the typist — and the dashboard is leaving your desk Cursor released a mobile app this week that lets developers steer coding agents from a phone, away from the editor.

2. The Semgrep team reused its frontier-agent test on open models. A Chinese one beat Claude. Semgrep had a narrow question, not a leaderboard to settle.

3. The health details you type into ChatGPT are legal to sell. Lawmakers want to change that. Two Democratic lawmakers are preparing a bill to bar the sale of Americans' health and location data to brokers, and its reach extends somewhere users rarely consider: the messages they type into AI

In Brief

• South Korea's chipmakers pledge $550B to expand memory production Samsung and SK Hynix, the world's two largest memory makers, committed over $550 billion to build additional fabs and memory research facilities. The spending responds to a severe DRAM and HBM shortage straining AI hardware supply, locally dubbed "RAMageddon."

• Anthropic gives California government Claude at half price Anthropic signed a deal with Governor Newsom letting California state agencies buy Claude access at a 50% discount. The agreement deepens Anthropic's ties to the state while it remains at odds with the federal government.

• Meta contractors posed as teens to test rival chatbots Hundreds of contractors on a Meta project impersonated minors to probe how Gemini and ChatGPT respond to prompts about suicide, sex, and drugs, WIRED found. The testing aimed to map competitors' safety responses on high-risk topics.

• HP expands OpenAI partnership across its operations HP scaled its OpenAI Frontier partnership to deploy AI in customer support, internal software development, and enterprise operations. The deal extends OpenAI's reach into a major PC manufacturer's product and back-office systems.

• Arena turns its free AI leaderboard into a $100M business Arena, the crowd-voted model-ranking site widely cited across the industry, now runs a $100 million commercial operation. The startup launched its paid service in September 2025.

• Palantir builds a secure AI engine for US agencies on NVIDIA Nemotron Palantir released an inference engine running NVIDIA Nemotron open models inside closed government environments. The system targets US agencies that need to run AI on classified data without external connectivity.

• Anthropic's Claude reaches general availability on Azure GB300 hardware Microsoft made Claude models in its Foundry platform generally available, hosted on Azure and running on NVIDIA GB300 Blackwell Ultra GPUs. Azure enterprises can now build agents on Claude without leaving Microsoft's stack.

• Flexion trains a humanoid robot for office tasks Flexion Robotics, founded by former Nvidia engineers, demonstrated a humanoid robot handling routine office work. The company uses a training method aimed at teaching robots practical manual tasks.

• Proception settles Tesla trade-secret suit and raises $11M Robot-hand startup Proception resolved Tesla's trade-secret lawsuit and announced an $11 million raise. The company collects training data to build dexterous robotic hands, one of robotics' hardest problems.

• TIDAL stops paying for AI-generated music TIDAL cut monetization for AI-generated tracks and will use automated tools to remove uploads that impersonate real artists or groups. The policy targets royalty payouts on synthetic music.

• Google opens Gemini's personalized image generation to free US users Google extended Gemini's personalized image generation to eligible free US accounts. The feature creates images using a user's interests and data pulled from connected Google apps.

• Firefly runs NVIDIA Jetson in lunar orbit Firefly Aerospace's Blue Ghost Mission 2, targeted for late 2026, will run its Ocula imaging service on NVIDIA Jetson in lunar orbit. The system processes images on orbit and transmits only selected results to Earth.

Read the full edition →