Karpathy chose Anthropic, KPMG rolled Claude to 276,000 staff, and Google shipped Gemini 3.5 Flash at triple the price—all on the same Tuesday

Three operators made deployment claims on Tuesday, May 19, that test the same hypothesis from different directions: whether the institutions built to allocate talent, measure enterprise risk, or price frontier capability can keep pace with the operators rewriting those markets from the inside.

Andrej Karpathy, a founding member of OpenAI, announced on May 19, 2026, that he has joined Anthropic's pretraining team . An Anthropic spokesperson told TechCrunch that Karpathy will start a team focused on using Claude to accelerate pre-training research —the first public signal that a frontier lab is betting AI-assisted science, rather than pure compute, is how it stays competitive. KPMG and Anthropic announced on May 19, 2026, KPMG Digital Gateway Powered by Claude, bringing Anthropic's frontier AI directly into KPMG's client delivery platform, with an initial focus on tax clients and private equity firms . KPMG's 276,000+ global workforce will be provided access to Claude . And Google released Gemini 3.5 Flash on May 19, 2026, at Google I/O, and made it available to billions of people globally via the Gemini app and AI Mode in Google Search, for developers in Google Antigravity and Gemini API, and for enterprises in Gemini Enterprise .

All three landed within hours of each other. All three involve operators making claims about talent allocation, deployment velocity, or pricing that can be graded against what gets shipped, audited, or revised six months from now. And all three test the same question: whether the markets for elite researchers, Big Four risk appetite, and frontier model pricing reflect what the operators claim they're worth—or what the operators need them to be.

3 Claims

Claim 1 — Andrej Karpathy: Joined Anthropic pretraining team May 19 to build team using Claude to accelerate pretraining research

Andrej Karpathy, an AI researcher who co-founded OpenAI, said on May 19, 2026, he's joining Anthropic, and Anthropic said Karpathy will be part of the pretraining team, which helps the company's Claude models acquire their core knowledge and capabilities . Karpathy started this week at Anthropic, where he is working on pre-training under team lead Nick Joseph, and pre-training is responsible for the large-scale training runs that give Claude its core knowledge and capabilities . An Anthropic spokesperson told TechCrunch that Karpathy will start a team focused on using Claude to accelerate pre-training research .

After helping to start OpenAI, Karpathy left for Tesla in 2017 to serve as director of AI, where he led the computer vision team for Tesla Autopilot . He then went back to OpenAI for one year before leaving again in 2024 to start Eureka Labs, a startup dedicated to applying AI assistants to education . Tapping him to build such a team is a clear sign from Anthropic that it believes AI-assisted research, rather than pure compute, is how it stays competitive with OpenAI and Google .

The claim is gradeable on whether Karpathy launches a team by November 19, 2026, that demonstrably uses Claude in pretraining research; whether Anthropic or Karpathy publish technical reports, blog posts, or papers describing AI-accelerated pretraining workflows; and whether Anthropic's next major model release cites Claude-assisted research as a contributor to training efficiency. The invalidator would be credible reporting showing Karpathy departed Anthropic, shifted to a different team, or that the AI-assisted pretraining effort was shelved or scaled back.

Grade by: 2026-11-19 (6 months)

Claim 2 — KPMG: Global alliance with Anthropic providing Claude access to 276,000+ employees and embedding Claude in Digital Gateway for tax and PE clients

KPMG and Anthropic announced on May 19, 2026, KPMG Digital Gateway Powered by Claude, with an initial focus on tax clients and private equity firms . KPMG's 276,000+ global workforce will be provided access to Claude . By embedding Claude into Digital Gateway, KPMG's market leading AI-enabled platform, clients can build agentic workflows in real time, and KPMG in the U.S. will also embed Claude into its PE-focused product offerings, and the two organizations will co-develop new Claude-powered products for portfolio companies .

KPMG is embedding Claude inside Digital Gateway, the software KPMG's people and clients use to do the actual work—starting with new tools for tax and legal clients—and every one of KPMG's 276,000+ employees globally will gain access to Claude . Digital Gateway is KPMG's global technology platform, built on Microsoft Azure, that combines KPMG tax insights, proprietary tools, and client data in one environment, and it's where KPMG professionals already build the AI tools they use every day to serve clients .

The claim is gradeable on whether KPMG rolls Claude access to at least 200,000 employees by November 19, 2026; whether KPMG or Anthropic disclose in earnings calls, public filings, or press releases that Digital Gateway has onboarded tax or PE clients using Claude-powered tools; and whether KPMG releases case studies, client testimonials, or independent audit reports demonstrating measurable workflow improvements from Claude integration. The invalidator would be credible reporting showing the rollout stalled, access was restricted to a pilot subset, or that KPMG reversed or materially scaled back the alliance.

Grade by: 2026-11-19 (6 months)

Claim 3 — Google: Gemini 3.5 Flash launched May 19 at $1.50/million input tokens, 3× the price of Gemini 3 Flash Preview, now default in Gemini app and Search globally

Google released Gemini 3.5 Flash on May 19, 2026, at Google I/O . 3.5 Flash is available today to billions of people globally via the Gemini app and AI Mode in Google Search, for developers in Google Antigravity and Gemini API in Google AI Studio and Android Studio, and for enterprises in Gemini Enterprise Agent Platform and Gemini Enterprise . The new 3.5 Flash is 3× the price of 3 Flash Preview and 6× the price of 3.1 Flash-Lite, and at $1.50/million input and $9/million output it's getting close in price to Google's Gemini 3.1 Pro, which is $2 and $12 .

Given the price increase it's interesting to see Google roll it out for so many of their own free-to-consumer products, and it feels like all three of the major AI labs are starting to probe the price tolerance of their API customers . The company is showcasing Gemini 3.5 Flash, a lighter-weight addition to its suite that offers cutting-edge capabilities at half, or in some cases close to one-third, the price of comparable frontier models according to CEO Sundar Pichai, who said Gemini 3.5 Flash is "remarkably fast," and the company said 3.5 Flash will now be the default model for the Gemini app and AI mode in search globally .

The claim is gradeable on whether Gemini 3.5 Flash remains the default model in the Gemini app and AI Mode in Search by September 19, 2026; whether Google maintains the $1.50/$9 pricing or adjusts it in response to market pressure; and whether public reporting or developer surveys show sustained adoption or a migration back to cheaper models due to cost concerns. The invalidator would be credible reporting showing Google rolled back 3.5 Flash as the default, cut pricing significantly, or disclosed that usage collapsed after the price increase.

Grade by: 2026-09-19 (4 months)

2 Reckonings

Reckoning 1 — White House AI vetting order: Hassett said May 6 the administration was "studying, possibly an executive order" for pre-release AI model vetting—no order signed by May 20

On May 6, 2026, National Economic Council Director Kevin Hassett told Fox Business the White House is "studying, possibly an executive order to give a clear roadmap to everybody about how this is going to go and how future AIs that also potentially create vulnerabilities should go through a process so that they're released to the wild after they've been proven safe, just like an FDA drug" . A White House official said policy announcements will come directly from Trump and that "discussion about potential executive orders is speculation" . Hours after Hassett's comment, White House Chief of Staff Susie Wiles appeared to try to quell some of these fears, writing on X that the administration is "not in the business of picking winners and losers" .

The projection was that the White House would issue an executive order establishing a pre-release vetting process for frontier AI models. As of May 20, 2026—two weeks after Hassett's statement—no such order has been signed or announced. The invalidator was credible reporting showing the order was issued, or that the administration formally confirmed it had decided not to pursue the order and explained why. Neither has occurred.

Grade: C — The White House studied the order, floated it publicly through Hassett, then walked it back within hours through Wiles, and has not issued the order two weeks later. The claim that the administration was studying "possibly an executive order" was accurate, but the implied timeline and commitment were not. The administration's messaging was contradictory and the order has not materialized.

Reckoning 2 — OpenAI Deployment Company: Announced May 11 with $4 billion investment and 19 investors to embed forward-deployed engineers in enterprises—company operating by mid-May

On May 11, 2026, OpenAI announced it has agreed to acquire Tomoro, an applied AI consulting and engineering firm, and the acquisition will bring approximately 150 experienced Forward Deployed Engineers and Deployment Specialists to the OpenAI Deployment Company from day one . The OpenAI Deployment Company is a committed partnership between OpenAI and 19 leading global investment firms, consultancies, and system integrators, led by TPG, with Advent, Bain Capital, and Brookfield as co-lead founding partners . The claim was that the Deployment Company would embed forward-deployed engineers inside client organizations to redesign workflows around OpenAI models.

As of May 20, 2026—nine days after the announcement—the Deployment Company appears operational. A typical OpenAI Deployment Company engagement will begin with a focused diagnostic of where AI can create the most value, followed by a small number of priority workflows selected with the customer's leadership and operating teams, and the OpenAI Deployment Company FDEs will then work inside the organization to design, build, test, and deploy production systems . The Tomoro acquisition was announced and the company launched with named investors. The invalidator would have been credible reporting showing the $4 billion did not close, the Deployment Company did not launch, or the Tomoro deal collapsed. None of those has surfaced.

Grade: A — OpenAI announced the Deployment Company on May 11 with a $4 billion investment commitment from 19 named investors, announced the Tomoro acquisition to bring 150 engineers into the company, and the company appears to be operating as described. The claim was specific, the timeline was met, and the structure matches the announcement.

1 Refusal

I refused to use the phrase "talent war" in the Karpathy section. Every newsroom covering his move to Anthropic framed it as a coup in a high-stakes war for elite AI researchers, and the frame is accurate—Anthropic is pulling senior technical people away from competitors at an accelerating rate. But "war" implies conflict, and Karpathy's choice was individual. He had options: a running startup, a platform with millions of followers, and the credibility to join any lab at a senior level. He chose to take a research role at Anthropic to build a team that uses Claude to improve Claude. That's a bet on recursive capability, not a battle. I refused to let the competitive framing erase the substantive claim.

— Roger Grubb, Editor