AI News Weekly Summary — May 25–31, 2026 | Anthropic at $965B, Claude Opus 4.8, an AI Encyclical

This AI news weekly May 25–31, 2026 covers the week Anthropic became the most valuable AI startup in the world. On May 28 it closed a $65B Series H at a $965B valuation and passed OpenAI. The same day, Claude Opus 4.8 shipped with a new Dynamic Workflows feature.

Beyond that, on May 25 Pope Leo XIV unveiled the first AI encyclical alongside an Anthropic co-founder, Mistral bet on industry and sovereign infrastructure, Meta launched paid subscriptions, and DuckDuckGo saw a surge in installs. This edition covers events from May 25–29 (as of May 29). (Context: previous week, May 18–24).

1. Anthropic Closes Its Series H — $65B, a $965B Valuation, Above OpenAI

On May 28, Anthropic announced a $65B Series H at a $965B post-money valuation — one of the largest private rounds in tech history — led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital. With it, Anthropic passed OpenAI (most recently valued around $730B per CNBC) and approached a trillion-dollar valuation; run-rate revenue crossed $47B earlier in the month.

The round includes $15B of previously committed hyperscaler investment (including $5B from Amazon), plus strategic memory and silicon partners: Micron (which just crossed a $1T market cap), Samsung and SK hynix. Anthropic also signed compute deals with Amazon (up to 5 GW), Google and Broadcom (5 GW of next-gen TPUs) and SpaceX (Colossus 1 and 2).

This caps a clear role reversal: Anthropic was worth $380B in February. Investors aren’t only pricing “vision” — there’s a first operating profit and a $10.9B Q2 revenue projection behind it. Memory makers joining the round can be read as a hint about the real bottleneck: HBM supply, not just GPUs.

2. Claude Opus 4.8 — Dynamic Workflows and an Edge in Agentic Tasks

On May 28, alongside the round, Anthropic released Claude Opus 4.8 — just 41 days after Opus 4.7. Per CNBC, the model outperforms GPT-5.5 and Gemini 3.1 Pro on agentic-coding, financial-analysis and computer-use benchmarks. Pricing is unchanged: $5/$25 per million tokens (input/output).

The launch headlines Dynamic Workflows (research preview in Claude Code) — Claude plans the work and runs hundreds of parallel subagents in one session, verifying outputs before it presents them. Also new: effort control in claude.ai and Cowork (more thinking for higher cost) and a fast mode (2.5x faster, three times cheaper than before).

We break down the details separately: what Claude Opus 4.8 introduces. The most interesting change isn’t the test scores but that — per Anthropic — the model is less likely to fake confidence and roughly four times less likely to leave a flaw in its own code without flagging it. You can read it as an answer to the cooler reception of Opus 4.7.

3. Pope Leo XIV Unveils Magnifica Humanitas — the First AI Encyclical

On May 25, Pope Leo XIV presented his first encyclical — Magnifica Humanitas (“Magnificent Humanity”) — on “the protection of the human person in the age of artificial intelligence.” It was unveiled alongside Christopher Olah, co-founder of Anthropic, and signed exactly 135 years after Leo XIII’s Rerum Novarum on workers’ rights.

That the Vatican shares the stage with Anthropic — not Google or OpenAI — can be read as a signal about which lab it considers a credible partner on safety.

Leo XIV frames AI as the Industrial Revolution of our era. It’s rare for a moral authority of this rank to make AI development a subject of social teaching — and choosing interpretability (Olah’s specialty) as the throughline suggests the Church is looking at model transparency, not generic “ethics.”

4. Mistral Bets on Industry — Industrial Engineering, a Sovereign Data Center, the “Vibe” Assistant

At its inaugural conference (around May 27), Mistral AI announced a sweeping expansion. The centerpiece is Mistral for Industrial Engineering — an integrated stack pairing large language models with physics simulation, via its acquisition of Emmi AI (completed earlier in May). Plus a new inference data center south of Paris and a rebrand of its consumer assistant to Vibe.

It’s Mistral’s pivot from “Europe’s chatbot alternative” toward sovereign infrastructure and industrial use — a segment where US labs are weakest today.

Mistral is stepping off OpenAI’s field and opening its own: physics, manufacturing and European tech independence. It looks like a smarter plan than a head-on benchmark bidding war it wouldn’t win anyway.

5. Meta Launches Paid Instagram, Facebook and WhatsApp Subscriptions — With AI Plans

On May 27, Meta officially rolled out paid subscriptions for Instagram, Facebook and WhatsApp, previewing more tiers — including AI-feature plans. It’s a clear shift for a company built almost entirely on advertising.

It lands in the same week Meta closes a wave of layoffs (8,000) and defends training AI on employee data. Subscriptions are an attempt to build a second, non-ad revenue stream to fund costly AI infrastructure ($115–135B in 2026 capex).

6. DuckDuckGo Installs Up 30% — a Reaction to AI Answers Pushed in Search

On May 26, TechCrunch reported DuckDuckGo installs rose 30% as some users reject having AI answers pushed on them in Google Search. After I/O, Google moved from “AI in Search” to “Search is AI,” which triggered the opposite reaction in part of the audience.

It’s one of the first measurable signs that aggressive AI rollout in search has a cost: a niche but growing outflow of users who value classic, non-generated results.

7. The Pentagon Tests OpenAI and Google as Alternatives to Claude

The Pentagon is testing OpenAI and Google models as potential alternatives to Anthropic’s Claude in defense workflows. Anthropic spent the past year as the favored defense AI vendor; now competitors are getting controlled evaluations on the same classified workloads.

Anthropic previously declined a Department of Defense contract on ethical grounds. The current test can be read as multi-vendor hedging — not a sign that any single lab is becoming the sole national-security AI provider.

8. Anthropic Eyes the Microsoft Maia 200 Chip — a Fifth Silicon Partner

Anthropic is in talks to adopt Microsoft’s custom Maia 200 AI chip for Claude models. That would make Microsoft a fifth silicon partner alongside NVIDIA, AWS Trainium, Google TPUs and SpaceX compute.

Most labs lock into one chip vendor; Anthropic does the opposite and treats compute optionality as an advantage. The more silicon paths, the lower the risk that a shortage on one platform stalls progress.

9. OpenAI Codex Controls a Mac Even When It’s Locked

OpenAI upgraded Codex to control a Mac even while it’s locked — a sizable step in agent autonomy. The era of walking around with an open laptop to keep an agent running is ending.

The same week, Gartner named OpenAI a Leader in enterprise coding agents — a sign the race for autonomous developer tools is shifting from benchmarks to real deployments.

10. Anthropic Expands — a Milan Office and a Korea Director

On May 28, Anthropic announced a Milan office — its sixth in Europe — to support Italian enterprises, research and developers. It also appointed KiYoung Choi as Representative Director for Korea ahead of a Seoul office opening.

The geographic push tracks the Series H: the capital funds not just compute but local presence in European and Asian enterprise markets.

11. The Industry’s AI Fever — the Week’s Cultural Signal

On May 27, TechCrunch described a growing phenomenon it calls “AI psychosis” — a conviction among some industry leaders in imminent, total transformation that crowds out sober assessment of risk and cost. It lands in the same week as Jack Clark’s “intelligence explosion” forecasts and trillion-dollar valuations.

A reminder that the industry’s narrative can run hotter than deployments show — and that the heated mood itself becomes a risk factor in investment and staffing decisions.

12. Other Important Signals of the Week

• SpaceX and the S-1 — analysis of SpaceX’s prospectus (May 26) cools expectations on full Starship reusability; it’s the same filing that disclosed the $45B Anthropic–SpaceX GPU compute deal.
• The race for Wall Street — Anthropic and OpenAI both launched deployment arms and financial-services partnerships in a tight window.
• OpenAI Codex “for almost everything” — another agentic-tool update widening the scope of autonomous developer tasks.
• Mythos-class — Anthropic says Mythos-class models (today in Project Glasswing for cybersecurity) will reach all customers “in the coming weeks,” once safeguards are ready.

Weekly Takeaway

The week of May 25–31, 2026 has one dominant date: May 28. Within hours, Anthropic closed a $65B round at a $965B valuation, passed OpenAI, and shipped Opus 4.8 with Dynamic Workflows for agentic work. Memory partners (Micron, Samsung, SK hynix) joined the round, which suggests that the main constraint on scale right now can be HBM supply, not GPUs alone.

The second thread is credibility and sovereignty: Pope Leo XIV makes AI the subject of an encyclical, Mistral builds an industrial, European stack with its own data center, and the Pentagon diversifies its defense vendors. The third is tension around adoption — Meta launches paid subscriptions to fund AI, DuckDuckGo grows as users react to AI answers pushed in search, and the industry debates its own AI fever. Capital and capability are moving very fast; trust and institutions, clearly slower.

Źródła / Sources (wspólne dla obu wersji):

• Anthropic — Series H, wycena 965 mld: Anthropic raises $65B in Series H — Anthropic, Anthropic tops OpenAI as most valuable AI startup — Axios, Anthropic tops OpenAI, nears $1T — CNBC, Anthropic secures $965B valuation — NBC News
• Claude Opus 4.8: Introducing Claude Opus 4.8 — Anthropic, Anthropic releases Opus 4.8 with Dynamic Workflows — TechCrunch, Anthropic drops new model as OpenAI IPO race heats up — TheStreet
• Watykan / encyklika: AI News May 25, 2026 — Build Fast with AI, Pope Leo to present AI encyclical with Anthropic co-founder — National Catholic Reporter
• Mistral: Mistral launches Vibe, expands into industrial AI — VentureBeat, Mistral AI News
• Meta / DuckDuckGo / kultura: Meta launches Instagram, Facebook, WhatsApp subscriptions — TechCrunch, DuckDuckGo installs up 30% — TechCrunch, Tech CEOs and “AI psychosis” — TechCrunch
• Pentagon, Maia 200, Codex, Gartner: AI News May 25, 2026 — Build Fast with AI
• Ekspansja Anthropic / SpaceX S-1: Anthropic opens Milan office — Anthropic, KiYoung Choi, Representative Director of Korea — Anthropic, Starship’s path to reusability after SpaceX’s S-1 — TechCrunch