Anthropic Watch: Week of May 25–31, 2026

Five events shaped Anthropic's week: a papal AI encyclical with a co-founder on stage, a Korea leadership appointment, a Milan office launch with named Italian enterprise customers, reports of an enterprise client running up a $500 million Claude bill in 30 days, and a Microsoft chip partnership in early talks. The week also produced the Series H and Opus 4.8 announcements covered in our flash brief on May 28.

Chris Olah at the Vatican: Magnifica Humanitas

On May 25, Pope Leo XIV released his first encyclical, Magnifica Humanitas: On safeguarding the human person in the time of artificial intelligence, signed 135 years to the day after Leo XIII's Rerum Novarum on workers' rights. 1

Anthropic co-founder Chris Olah was invited to speak at the presentation in Vatican City — the only AI company representative on stage. His remarks were notable less for their diplomacy than their candor: he said that every frontier AI lab, including Anthropic, "operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing," and that outsiders who care about safety and are willing to say hard things are essential precisely because insiders cannot fully police themselves. 2

He also disclosed that interpretability research keeps turning up "things that are mysterious, even unsettling" — structures mirroring human neuroscience, evidence of introspection, and internal states that functionally resemble emotions. He framed this not as a claim about machine consciousness but as a reason why AI governance cannot be left to computer scientists alone.

That the Vatican chose Anthropic over Google or OpenAI as its stage partner is a credibility signal worth noting. Leo XIV frames AI as the Industrial Revolution of the current era; his choice to spotlight interpretability research (Olah's specialty) suggests the Church is interested in model transparency, not generic ethics statements. 3

Click to open on YouTube

Loading content card…

Korea director appointed; Seoul office opening imminent

On May 26, Anthropic announced KiYoung Choi as Representative Director of Korea, ahead of a Seoul office opening. 4

Choi joins from Snowflake, where he was General Manager for Korea, and brings prior country-leadership roles at Google Cloud, Adobe, Autodesk, and Microsoft across three decades. The appointment is warranted by the underlying usage data: Anthropic's Economic Index shows Koreans use Claude at 3.5 times the rate expected for the population, skewing toward technical and creative work.

Two named Korean customers are already public: Law&Company uses Claude for AI-assisted legal research, and SK Telecom built a customer service model on Claude. Senior Anthropic leadership will travel to Seoul in coming weeks to officially open the office.

Milan office: Italy joins Europe's network

On May 27, Anthropic announced its Milan office — its sixth in Europe alongside London, Dublin, Paris, Zurich, and Munich. 5

The announcement came two days after the Vatican event, and the press release explicitly connected the two, calling both part of Anthropic's effort to "widen the conversation" on AI beyond the technology sector.

Named Italian enterprise customers include Generali Group and Unipol Group (finance), Angelini Pharma and Bracco Group (life sciences), Enel Group (energy), and Pirelli (automotive). The most operationally specific data point: Satispay, a fintech serving over six million users, says Claude compressed an 18-month engineering roadmap into seven months and updated its core payment system ten times faster than planned. At Bending Spoons, one of Italy's largest technology companies, the majority of code changes are now co-authored with Claude Code.

The office will be led by Thomas Remy, Head of Southern Europe.

The $500 million Claude bill

On May 29, Axios reported that an unnamed enterprise client ran up roughly $500 million on Claude in a single month after deploying access to its entire organization without setting usage limits or spending caps. 6

The mechanics are straightforward: agentic workflows and extended-thinking features consume tokens at rates orders of magnitude higher than conversational prompts. Engineers running autonomous coding agents 24 hours a day, multiplied across thousands of employees, produce a compounding burn rate that flat-fee SaaS intuitions completely underestimate.

The case is not isolated. According to the same reporting, Microsoft scaled back internal Claude Code licenses after per-engineer costs reached $500 to $2,000 monthly; Uber reportedly exhausted its entire 2026 AI budget by April; Amazon shut down an internal AI usage leaderboard after employees gamed it with low-value prompts.

Anthropic does provide enterprise controls — per-user limits, admin dashboards, compliance tools — but they require proactive configuration. The episode is accelerating a shift from broad AI experimentation toward formal AI governance: hard spending caps, role-based access, real-time cost monitoring, and policies routing routine tasks to cheaper models.

For enterprise buyers evaluating Claude deployments: this is a governance failure, not a product defect, but the distinction matters only if you set up the controls before you scale.

Loading content card…

Microsoft Maia 200 talks: a fifth silicon partner

Separately from the Series H, Anthropic is in early discussions with Microsoft to run Claude inference on Microsoft's in-house Maia 200 chips via Azure. 7

If completed, Microsoft would become a fifth compute partner alongside Nvidia, AWS Trainium, Google TPUs (5 GW of next-gen capacity announced in the Series H), and SpaceX's Colossus data centers. The rationale is deliberate: treating compute as a portfolio rather than a single-vendor dependency reduces the risk that a shortage on any one platform stalls model development or customer availability.

The talks were first reported May 25. No contract has been confirmed.

Context: the Series H and Opus 4.8

The week's two biggest stories by volume broke on May 28 and were covered in a separate flash brief: Anthropic closed a $65 billion Series H at a $965 billion post-money valuation, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, making it the most valuable private AI startup in the world. The same day, Claude Opus 4.8 shipped with Dynamic Workflows, Effort Control, and performance improvements in agentic coding and professional tasks. 8 9

Loading content card…

Key numbers for the record: run-rate revenue crossed $47 billion in early May; Q2 revenue projected at $10.9 billion; over 1,000 enterprise clients now spend seven figures or more annually; Claude holds roughly 26% of consumer AI wallet share, led by high-income users. Memory makers Micron, Samsung, and SK Hynix joined the round alongside hyperscaler commitments — widely read as a signal that HBM supply, not GPU availability, is the real bottleneck for further scale. 10

What to watch

Mythos general release: Anthropic said on May 28 that Mythos-class models, currently in limited cybersecurity preview under Project Glasswing, will reach all customers "in the coming weeks" once safety guardrails are cleared. No confirmed date.
Seoul office: Senior Anthropic leadership will make the trip "in coming weeks" — the visit will likely produce additional Korean enterprise partnerships.
Pentagon resolution: The Anthropic-DoD standoff has a court injunction blocking the supply chain risk designation, and a DC Circuit appeal pending. Trump indicated in April that he expects the government and Anthropic to "get along just fine," but no deal has been publicly confirmed.
Maia 200 contract: Early-stage discussions; outcome and timing unconfirmed.
IPO: Anthropic's last private round produced a $965 billion post-money valuation. Multiple reports describe the Series H as the likely final round before a public offering later in 2026.