1. OpenAI Shipped GPT-5.6 in a Day. The Government Decides Who Gets In. The Trump administration told OpenAI to slow down. The reason it gave was safety.

2. Checking a coding agent's work is now the hard part, not writing the code For most of computing's history, one assumption held steady: confirming a solution is easier than producing it. A paper circulating on Hugging Face says that for coding agents, the order has reversed.

3. The week creative software stopped treating AI as a side panel Two product announcements landed days apart, from companies that share no roadmap. Figma used its annual Config conference to fold AI into the work itself, not bolt it on.

In Brief

• Anthropic accuses Alibaba of mining Claude through 25,000 accounts Anthropic asked regulators to penalize Alibaba, alleging the company ran 28.8 million exchanges across 25,000 accounts to extract Claude's capabilities. Anthropic frames it as the largest model-cloning attempt against its systems.

• NYT amends copyright suit to target Microsoft's supercomputer The New York Times revised its copyright claims against OpenAI and Microsoft, arguing Microsoft built the training supercomputer that enabled infringement. The NYT shifted its theory after a Supreme Court ruling against Sony.

• Apple cancels high-end M6 Mac chips for an AI-focused M7 line Apple will skip the high-end M6 tier and ship M7 Pro, M7 Max, and M7 Ultra instead, prioritizing AI workloads. The shift reorders Apple's silicon roadmap for Mac.

• IBM claims first sub-1-nanometer chip technology IBM announced nanostack transistors it says break the 1-nanometer threshold. IBM says the design can raise chip performance or cut energy use, though it remains a research milestone rather than a shipping product.

• OpenAI hires Uber's India chief to run its largest market outside the US OpenAI recruited Uber India's head to lead its India operations, its biggest market beyond the United States. The company is expanding offices, partnerships, and hiring there.

• South Korea will train its entire military on drones South Korea plans to train its roughly 500,000-strong military to operate drones as a standard combat tool. The program treats drone skills as a universal requirement across all units.

• Netris raises $15M Series A from a16z for neocloud networking Netris closed a $15 million Series A led by a16z. Its software runs on network switches and cuts the time AI neocloud operators need to bring capacity online.

• TechCrunch argues AI policy now needs collective action beyond a two-lab rivalry TechCrunch contends model capabilities now carry direct political consequences that the OpenAI-versus-Anthropic framing no longer captures. The piece calls for industry-wide coordination on the resulting risks.

• Researchers propose component-level benchmarks for LLM agent memory A new paper argues agent memory has grown into a full data-management system handling storage, retrieval, updates, and consolidation. The authors say end-to-end metrics like F1 hide failures and call for testing memory components directly.

• Qwen-Image-Agent adds planning and search to text-to-image generation Alibaba researchers built an agentic framework that plans, reasons, searches, and uses memory to fill gaps in underspecified image prompts. They target requests that need current knowledge or implicit context standard text-to-image models miss.

Read the full edition →