As of August 11, Anthropic embeds an invisible machine-readable watermark in all Claude-generated text and C2PA provenance metadata in generated images, across Claude.ai, the API, Claude Code, and Claude on AWS, Google Cloud, and Microsoft Foundry. The mark survives copy, paste, and light edits; heavy paraphrasing or translation degrades it. Why it matters. This is infrastructure-level, not a consumer toggle. Universities, publishers, and enterprise legal teams that had no way to detect Claude output now will, and a third-party detection API is on the way. TechCrunch has the scope →

Nvidia partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to stand up independent platforms targeting more than $500 billion of third-party capital, treating GPU infrastructure as a lendable asset class akin to commercial real estate. Customers borrow against the hardware rather than paying upfront; the six managers pick the projects, and Nvidia does not run the pools. Why it matters. It removes the capital wall that had been slowing enterprise-scale deployments, though a chip vendor arranging financing for its own buyers will draw a hard regulatory look. CNBC →

Riot Platforms will supply 191 MW from its Rockdale, Texas campus, phased to 96 MW by December 2027 and full capacity by June 2028; two extensions could push total value to $16.1 billion. It follows Anthropic's $10 billion Volta Infra deal on August 4. Why it matters. Crypto miners already hold the substations and grid connections AI labs would otherwise spend two to four years building, so expect Marathon, CleanSpark, and Core Scientific to field the same call. Bloomberg →

The self-funded tender, completed August 10, held the same $852 billion mark from the March raise and was open to current and former staff, per Bloomberg. No new capital, no IPO timeline. Why it matters. A flat valuation funded from OpenAI's own cash reads two ways: it expects a higher mark in a public offering, or it feels no near-term fundraising pressure. Either is worth watching pre-IPO. Our writeup →

Brad Lightcap, COO for four years since joining in 2018, confirmed his departure on August 11 after moving to special projects in April. Chloe Bakalar, OpenAI's head of ethics and sole dedicated ethicist, also exited; the company says ethics duties are now spread across teams. Why it matters. Two governance departures at once, operations and ethics, ahead of a widely expected IPO is exactly the pattern institutional investors and regulators read for internal disagreement about direction. CNBC →

Announced August 10, Daybreak now splits in two tiers: Blue gives verified defenders GPT-5.6 Sol with threat-analysis prompting, and Red adds GPT-5.6-Cyber for exploit development, vulnerability research, and adversary simulation. Per OpenAI, the Cyber model answers 95% of advanced security prompts that standard Sol declines, against a 1.5% Sol baseline. Hardware security keys become mandatory for all tiers on September 1. Why it matters. That 95% versus 1.5% gap is the distance between what verified defenders can now do and what attackers already prompt out of less-aligned models. Our writeup →

Yelp wired Reservations and Waitlist into ChatGPT on August 10, with OpenTable and Resy also integrated, across the US and Canada on every plan tier. Find-to-book now happens in one conversation. Why it matters. This is the jump from research tool to operating layer, and the GEO consequence is direct: a business has to be in Yelp's data to surface in ChatGPT's booking flow. Flights and hotels are the obvious next verticals. Search Engine Land → Want the booking alerts routed into your own stack? Wire it up with Make →

The SpaceX acquisition closes this quarter, so we made the call for three buyers: solo devs, small teams, and enterprise. Read the verdict → Try Cursor →

A scored RFP template your security and finance teams can sign off on, so the evaluation is a rubric instead of a vendor demo. Get the scorecard →

Each of the three opens a distinct leak path; this walks all of them through a DLP lens so you know what to lock down first. Read the guide →

We pulled the six patterns that recur across every named public enterprise agent deployment this year, with the tradeoff each one buys you. See the patterns →