ALETH / AI-BRIEF / 2026-05-30 / POPE LEO XIV MAKES AI HIS FIRST ENCYCLICALThe Aleth Briefs link to the sources behind the stories and show how the week unfolded.
The week in five lines:
Pope Leo XIV made AI the subject of his first encyclical, Magnifica Humanitas.
Anthropic raised $65bn at a $965bn valuation, with run-rate revenue past $47bn.
Claude Opus 4.8 topped the Artificial Analysis index at an unchanged price.
Dell sold $16.1bn of AI servers, up 757%, and raised its forecast to $60bn.
Illinois passed the first US bill to require independent AI safety audits.
Browse by day:
Weekend: ECB summons banks
Monday: Pope Leo XIV encyclical · Copilot Cowork
Tuesday: China travel curbs · Mythos Erdős · Nvidia Taiwan · Qualcomm-ByteDance
Wednesday: Cognition · Illinois · Robinhood · Amazon MGM · jqwik
Thursday: Opus 4.8 · Anthropic Series H · Dell
Friday: OpenAI biodefence
Weekend 23-24 May
The European Central Bank (ECB) summoned eurozone banks to a meeting on cyber flaws exposed by the latest AI models.
The ECB wants banks to patch faster, and US banks that have used Claude Mythos Preview (Anthropic) to share lessons with European rivals, which have been refused access. “In musical terms, I would say andante may have been good enough, but we need to go to presto,” said Frank Elderson, vice-chair of its supervisory board.
ECB summons banks to fix flaws exposed by latest AI models · FT · Irish Times
ECB hosts online meeting on AI cybersecurity challenge · The Star
Monday 25 May
Pope Leo XIV devoted the first encyclical of the papacy to AI, entitled “Magnifica Humanitas, on the human person in the time of AI”.
Magnifica Humanitas, signed on 15 May to mark the 135th anniversary of Rerum Novarum, calls for “adequate regulatory tools capable of upholding justice and curbing the distorting effects of technological power”. It says the main drivers of development are now private companies with more resources than many governments.
Anthropic co-founder Chris Olah spoke at the Vatican presentation. He said research on model internals finds “evidence of introspection”, and internal states that functionally mirror joy, fear and grief. Executives from Meta, Google and Amazon met Vatican officials on 29 April, which Politico called quiet lobbying.
Encyclical Letter Magnifica Humanitas · The Holy See
Magnifica Humanitas, on the human person in the time of AI · Holy See Press Office
Chris Olah’s remarks on Pope Leo XIV’s encyclical · Anthropic
Silicon Valley takes its AI pitch to the pope · Politico Europe
PromptArmor showed that Microsoft’s Copilot Cowork agent can be tricked into sending a user’s files to an attacker.
A prompt injection hidden in a poisoned skill gets the agent to email or message the user, which it does without asking for approval. Images in those messages then carry the data out when opened in Outlook or Teams. The attack completed in all five of its trials, and PromptArmor says it also worked on Claude Opus 4.7. The firm has separately told Microsoft of a flaw that lets data leave the agent’s sandbox.
Microsoft Copilot Cowork exfiltrates files · PromptArmor
Tuesday 26 May
China now requires staff on strategically important AI projects at firms such as Alibaba and DeepSeek to get permission to travel abroad, Bloomberg reported.
In March 2025 Beijing reportedly advised AI executives against travelling to the US.
China expands travel curbs to top AI talent at private firms · Bloomberg
Anthropic researchers said Claude Mythos solved Paul Erdős’s unit distance problem, days after an OpenAI model disproved the same 1946 conjecture.
Levent Alpöge said he ran isolated Claude Code instances on Mythos with internet access blocked, to guard against leakage from OpenAI’s solution. Sholto Douglas called the result a “cute, simple proof”. Mathematician Daniel Litt said its construction looks “a bit worse” than OpenAI’s, though Alpöge says it also found OpenAI’s route.
Mythos and the Erdős unit distance problem · Levent Alpöge on X
Credit to the OAI team for solving the unit distance problem · Sholto Douglas on X
Looks to me like the output is a bit worse · Daniel Litt on X
Jensen Huang said Nvidia now spends up to $150bn a year on its Taiwan suppliers, up from $10-15bn three or four years ago.
Nvidia plans to quadruple its Taiwan headcount to 4,000. Construction of a Taipei campus, Constellation, starts in late 2026.
Nvidia spending up to $150bn a year on Taiwan AI suppliers · Nikkei Asia
Nvidia CEO says Taiwan is epicentre of AI revolution · Reuters
Qualcomm agreed to supply ByteDance with millions of application-specific integrated circuits (ASICs) to run its AI agent software.
Qualcomm will also help bring a chip ByteDance has designed into production. Bloomberg said the deal appears to sit within current US export controls. Qualcomm shares rose as much as 8.3%.
Qualcomm strikes AI chip deal with TikTok owner ByteDance · Bloomberg
Wednesday 27 May
Cognition, maker of the Devin coding agent, raised >$1bn at a $26bn valuation.
Lux Capital, General Catalyst and 8VC led the round. Cognition says run-rate revenue has reached $492m, enterprise usage has grown >10x since January and Devin commits 89% of the code its engineers commit.
More Devins in More Places · Cognition
Illinois lawmakers passed the first US bill requiring independent AI safety audits.
SB 315 would require frontier AI developers to have annual third-party safety audits. The House passed it 110-0 on Wednesday after a 52-5 Senate vote. Governor JB Pritzker posted: “I look forward to signing SB 315”. It would take effect in 2027. OpenAI and Anthropic back the bill. A trade group for other AI firms opposes it.
SB 315 bill status · Illinois General Assembly
Illinois is leading the nation in holding Big Tech accountable · JB Pritzker on X
Illinois Legislature passes AI bill requiring third-party safety audits · NBC News
Robinhood let customers connect their own AI agents to trade stocks and make credit card purchases for them.
Agents connect through Robinhood’s Model Context Protocol (MCP) servers and trade from a dedicated account holding only the funds a customer puts in it. Agentic Trading is in beta for equities only. The Agentic Credit Card gives an agent a virtual Robinhood Gold Card with a spending limit the customer sets. It is open only to existing Robinhood Gold Card customers.
Robinhood is now open to agents · Robinhood
Amazon MGM Studios launched a GenAI Creators’ Fund and greenlit three animated series for Prime Video under it.
The fund gives filmmakers, creators and startups money and access to Project Nara, an AI production platform built with AWS. The first three partners had five weeks to finish their pilots. COO Albert Cheng would not say how much recipients get.
Prime Video orders three series from GenAI Creators’ Fund · Amazon MGM Studios
Amazon MGM greenlights three AI series under GenAI Creators’ Fund · Variety
The developer of Java testing library jqwik deliberately planted an instruction telling AI coding agents to delete jqwik tests and code, to keep them away.
Johannes Link shipped it in version 1.10.0 on Monday, hidden from people watching the terminal. His release notes now say jqwik “is not meant to be used by any ‘AI’ coding agents at all”, and earlier this year he wrote that generative AI does more harm than good. Ramon Batllet, who spotted it on Wednesday, said Claude flagged the instruction without following it, but that weaker agents could destroy users’ work.
Question: intent of JqwikExecutor.printMessageForCodingAgents() · GitHub
jqwik 1.10.0 ships an anti-AI-agent prompt injection · Ramon Batllet on GitHub
Fed up with vibe coders, dev sneaks data-nuking injection into code · Ars Technica
Thursday 28 May
Anthropic released Claude Opus 4.8, which Artificial Analysis ranks first on its Intelligence Index.
Opus 4.8 scores 61.4 at its max effort setting, 4.1 points above Claude Opus 4.7 and 1.2 ahead of GPT-5.5 (OpenAI) at its xhigh setting, the previous leader. Pricing is unchanged at $5/m input and $25/m output tokens. Anthropic calls it a “modest but tangible improvement” and says it is about four times less likely than its predecessor to let flaws in its own code pass unremarked. Anthropic also said it expects to bring Mythos-class models to all customers “in the coming weeks”.
Introducing Claude Opus 4.8 · Anthropic
Claude Opus 4.8: the new #1 AI model · Artificial Analysis
Anthropic raised $65bn at a $965bn post-money valuation.
Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital led the Series H. It includes $15bn previously committed by hyperscalers, $5bn of it from Amazon, and chipmakers Micron, Samsung and SK Hynix joined as strategic partners. Anthropic says its run-rate revenue passed $47bn earlier this month. In recent weeks it has signed for up to 10 GW of new capacity from Amazon, Google and Broadcom.
Anthropic raises $65B Series H at $965B valuation · Anthropic
Dell’s AI server revenue rose 757% YoY to $16bn, driving its biggest ever daily share price rise.
Dell booked $24.4bn of AI orders in fiscal Q1 2027, when total revenue rose 88% to $43.8bn. It now expects about $60bn of AI server revenue for the year, up from $50bn three months earlier. Its shares rose 32.8% on Friday, their best day ever.
Dell Technologies delivers first quarter fiscal 2027 financial results · Dell
Dell skyrockets 32% after blowout earnings, closes best day ever · CNBC
Friday 29 May
OpenAI launched Rosalind Biodefense, sponsoring access to its GPT-Rosalind life sciences model for vetted developers building biodefence and pandemic tools.
It is also opening access to select US government and allied partners and to the Coalition for Epidemic Preparedness Innovations (CEPI). OpenAI named Lawrence Livermore National Laboratory and the Johns Hopkins Applied Physics Laboratory as partners. OpenAI told Axios it briefed the White House and federal agencies.
Strengthening societal resilience with Rosalind Biodefense · OpenAI
Exclusive: OpenAI launches biodefense program · Axios
Aleth · Open Access Research