← Dailies

Daily · 2026-09-18 00:00 UTC

OpenAI ships Astra for Law; Claude Code adds parallel cloud agents — 2026-09-18 00:00

OpenAI launches Astra for Law, Anthropic rebuilds Claude Code Projects around parallel cloud agents, Cloudflare blocks AI agents on ad-supported pages, and Crusoe raises $3.9B.

Source ↗

🎧 News Digest — 2026-09-18 00:00 UTC 10 items

1. OpenAI launches Astra for Law, a GPT-6 configuration for legal research — Astra for Law is not a new model: OpenAI wrapped GPT-6 Astra in a legal search index plus a set of instructions for legal analysis, and is offering the result to law firms and the vendors that sell to them. The index reaches US case law, statutes, regulations, court rules and administrative decisions across more than 230 million URLs, with sources added daily, and the Free Law Project's CourtListener database supplies the case law. It scored 54% overall correctness on 200 questions from the private validation set of Vals AI's Legal Research Bench, against 38.7% for GPT-6 Astra with web search alone at the same reasoning effort, and found 24% more reference cases on case-law questions. Access starts through a Trusted Access program in ChatGPT and Codex, where it appears as GPT-6 Astra Law; an API version is due later with no date or pricing attached, and Harvey and Legora are named as API customers. SiliconANGLE · Sep 17
2. Anthropic rebuilds Projects in Claude Code for parallel cloud agents — Users now describe a goal and a coordinator splits the work across parallel "threads", each running as its own cloud session that can open pull requests and run tests. Progress is trackable in the main chat or per thread, including on mobile, and Claude builds a shared memory across threads while a library collects uploaded files and results. The beta is open to select Pro and Max subscribers using cloud sessions, with Team and Enterprise access later and local execution coming soon. The Decoder · Sep 17
3. Anthropic publishes the metrics behind its call to pace the frontier — Days after chief executive Dario Amodei called on leading model makers to coordinate on slowing the pace of development, Anthropic detailed three metrics it is already measuring — AI-led research and development, oversight of autonomous AI agents, and compute allocation — and published the exact methodologies so other frontier labs can measure them too. "As the world considers pacing the frontier, we should do everything possible to minimize the gap between what frontier labs know and what the public knows," the company wrote. It argues compute is among the most verifiable inputs to the AI R&D process and could be a critical lever in a future pacing effort. SiliconANGLE · Sep 17
4. Cloudflare blocks AI agents by default on any page that carries an ad — Cloudflare's new preset for ad-supported domains allows search crawlers, refuses AI training, and blocks user-directed agents on any page that shows an ad, with chat-fetch bots and browser-use agents falling into that blocked category. There is no opt-out directive for agents yet, only a block. The default reaches new domains, new sites from existing customers and free-tier accounts. Cloudflare says AI training has grown to 52% of crawler traffic on its network, up from 22% last year, while fewer than 1% of its sites block search bots and 17% restrict AI training — the gap that prompted it to split one "Block AI Bots" switch into three controls. PYMNTS · Sep 17
5. Crusoe raises $3.9B at a $30.9B valuation for AI data centers — The Series F was co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, with Founders Fund, GIC, Nvidia, the Qatar Investment Authority, Radical Ventures and TPG also participating. The eight-year-old company will finance existing projects including a large site in Abilene, Texas used by OpenAI, plus smaller modular "AI factories" called Spark that are built in-house and can be trucked to large power sources — which also helps it sidestep local opposition to giant complexes. Crusoe leases space to customers who bring their own GPUs, rents out its own GPUs, and sells inference compute. TechCrunch · Sep 17
6. The UN turns to Google to make its global statistics readable by AI agents — The UN System Data Commons is built on Google's open-source Data Commons platform and lets people search statistics from across UN agencies using natural-language queries, replacing the browse-first UNData portal. It also supports the Model Context Protocol, so AI systems can connect to the data directly. The shift follows a UNICEF benchmark of six large language models across more than 133,000 responses on global development indicators, which produced an average accuracy score of just 21.2%; about three in five responses gave no usable number at all, and when the same questions were rerun about two days later, models that answered both times returned an identical number only about half the time. The study is a working paper not yet peer-reviewed. TechCrunch · Sep 17
7. Unsealed filings: a Microsoft director called AI scraping "theft of unprecedented proportions" — In court documents unredacted Thursday, lawyers for the New York Times in its copyright infringement case against Microsoft and OpenAI cited Microsoft's Director of Applied Science, Brent Hecht, writing that "millions of people around the world will soon consider large models 'hoovering up' all their work to be an astonishing theft of unprecedented proportions… almost no one intended for content they created to be used in this fashion, nor are they compensated for its use." The Orlando Sentinel, also part of the suit, reports a Microsoft filing framed Hecht as someone on the payroll to play a contrarian. The Verge · Sep 17
8. Meta ordered to remove UK deepfakes as its oversight board condemns "inadequate" safeguards — Meta's oversight board, whose decisions are binding, ruled that a fake video showing a Scottish Labour councillor making inflammatory comments about refugees should not have been left up on Facebook, and that an AI-generated video of a Muslim campaign volunteer should have been removed after falsely depicting her offering health advice while carrying out absurd exercises or eating junk food. The board said Meta's safeguards were "consistently and fundamentally inadequate" to address the rapid rise of AI deepfakes, and recommended policy changes including demoting "high risk" content in users' feeds. The Guardian · Sep 17
9. Australia considers a ban on smart glasses in Commonwealth offices — The Albanese government is weighing what it calls a "world-leading" ban on smart glasses in its offices and service centres, with councils around the country pushing the technology out of public spaces over privacy fears. The public service minister, Katy Gallagher, said she had asked the Australian public service for advice on whether the glasses should be prohibited, citing their "ability to record and capture information". Commonwealth-run workplaces range from bureaucrats' offices to Centrelink service centres and CSIRO laboratories, while state-run schools and hospitals would set their own rules. A University of Sydney study of smart-glasses footage posted to Instagram found 60% of a sample could be classified as potential harassment. The Guardian · Sep 17
10. A developer claims GPT-6 Astra cracked an 83-year-old Enigma message — A Bloomberg developer claims to have cracked a Wehrmacht Enigma message from 1941 that had resisted cryptanalysis for 83 years, using OpenAI's GPT-6 Astra. The 82-character radio message contains a soldier asking about his march route. The solution still needs independent review. The Decoder · Sep 17

🧪 Try This Next

• 🔧 Price out Astra for Law if you serve legal teams — the 230M-URL index and the Benchmark gap versus plain web search are the story
• 🎥 Read Anthropic's measurement methodology — if lab-level pacing rules arrive, these three metrics are the drafting template
• 📝 Check your crawler posture — Cloudflare's ad-page default now blocks user-directed agents with no opt-in directive yet

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.