The AI Brief

Vol. I · No. 3 · Thursday, May 28, 2026
The Vatican's first AI encyclical declares lethal autonomous weapons categorically impermissible — elevating the governance debate to 1.4 billion Catholics and beyond. The tech sector has shed more than 113,000 jobs in five months, with Meta's 8,000 cuts the most-watched single event. Google's Gemini Spark, a persistent cloud-resident agent, began reaching US subscribers this week. And Cohere's acquisition of Germany's Aleph Alpha — combined entity valued at $20 billion — is building a transatlantic sovereign AI stack that bypasses US infrastructure entirely.
One Thing

The Vatican names AI's lethal autonomy line — and holds it

Why it matters
On May 25, Pope Leo XIV released Magnifica Humanitas ("Magnificent Humanity"), a 42,300-word encyclical — the first papal teaching document devoted entirely to artificial intelligence — declaring it "not permissible" to entrust lethal decisions to autonomous systems, and calling for international AI governance on a scale comparable to nuclear arms control. Encyclicals carry doctrinal weight addressed not only to the Church's 1.4 billion members but, as this document explicitly states, to "every person of goodwill." The document is the most extensive magisterial treatment of AI to date, positioning the development of the technology as a social and anthropological challenge equivalent to industrialisation — the same framing Pope Leo XIII applied to labour and capital in Rerum Novarum in 1891. The Vatican holds no regulatory authority over private technology companies, but the historical precedent of Rerum Novarum — which reshaped labour law across multiple continents over subsequent decades — is precisely the precedent Leo XIV invokes. In an unprecedented break with Vatican tradition, Leo personally presented the document at the Synod Hall alongside Christopher Olah, co-founder of Anthropic, signalling that at least portions of the frontier AI industry are treating the encyclical as something beyond a symbolic gesture.
What's at stake
The encyclical enters a regulatory landscape where the EU AI Act, OECD AI Principles, and G7 statements are already competing for normative authority — none of them binding on autonomous weapons, which the EU AI Act explicitly excludes from its scope. The document has no enforcement mechanism. Its influence operates on soft-law timescales: in the roughly 40 countries where Catholic social teaching informs legislative culture — Poland, Italy, Brazil, the Philippines among them — it gives advocates a high-authority reference point in compliance and policy debates that did not exist on May 24. Whether that translates into binding treaty movement at the Geneva Convention on Certain Conventional Weapons — deadlocked since 2014 on autonomous weapons — is the structural question it leaves open.
Decode
Lethal autonomous weapons systems (LAWS) = military platforms that select and engage targets without a human making the final kill decision. They are legal under current international humanitarian law, which requires attacks to be distinguishable, proportionate, and subject to human judgment — a requirement LAWS satisfy algorithmically rather than by a person. No binding international treaty bans them despite more than a decade of Geneva negotiations. The encyclical categorises them as morally impermissible, resolving the debate on behalf of the Church rather than contributing to it.
Detail

Magnifica Humanitas was signed on May 15, 2026 — the 135th anniversary of Rerum Novarum — and formally released by the Holy See on May 25. The document spans 245 numbered paragraphs across five chapters, an introduction, and a conclusion, published in eight languages. On autonomous weapons, it establishes three explicit requirements: traceability of decisions, meaningful human control over lethal action, and international rules capable of slowing the technological arms race. On labor: the encyclical decries "new forms of slavery" among workers in AI supply chains and calls for worker protections equivalent to those debated during industrialisation. On governance: it demands democratic oversight of AI systems and calls for "robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility."

The document's presentation was itself a signal. Leo broke with centuries of Vatican tradition by speaking at his own encyclical's launch. Anthropic co-founder Christopher Olah — the researcher who built the field of mechanistic interpretability — shared the stage. Microsoft, IBM, and Cisco had previously signed the Vatican's Rome Call for AI Ethics; several of those same companies are now deeply embedded in US government AI infrastructure. The US has not formally responded to the encyclical's calls for binding international governance. The UN's AI advisory body, established in 2024, produces non-binding guidance. The Convention on Certain Conventional Weapons process in Geneva, where autonomous weapons have been debated since 2014, remains deadlocked.

Disclosure: Christopher Olah, who presented alongside the Pope at the encyclical's launch, is co-founder of Anthropic, the company that develops Claude, which generates this brief.

Vatican News — Magnifica Humanitas (primary)/ Time/ CNBC/ National Catholic Register — Full Text
Number of the Day
113,000+
Tech-sector jobs eliminated in 2026 through mid-May · 179 companies · avg. 825 per day · per Layoffs.fyi tracking

Record profits and record cuts — AI is repricing labour faster than it replaces it

Why it matters
The 2026 layoff wave has surpassed the post-pandemic correction of 2023 on a year-to-date basis, and the pace is accelerating: April alone accounted for more than 45,000 cuts — the worst single month in two years. What distinguishes this cycle from prior tech downturns is the explicit pairing of record financial performance with headcount reduction. Cisco cut 4,000 workers the same day it reported $15.8 billion in quarterly revenue. Meta began its 8,000 cuts on May 20 against record quarterly revenue of $56.3 billion. The money being freed from payroll is being redirected almost entirely to AI infrastructure: Google, Amazon, Meta, and Microsoft combined are targeting $725 billion in AI capital expenditure in 2026, up 77% year over year. The structural question is whether the AI jobs being created — estimated at 275,000 open roles — will absorb the displaced workforce at comparable wages.
What's at stake
For most operators, this is context, not a decision. For teams building on vendor platforms undergoing restructuring — Oracle, Cisco, Salesforce, Meta — vendor-risk due diligence now includes understanding whether the cuts are performance-based (transient) or structural (permanent reallocation toward AI). Bloomberg data suggests roughly half of AI-attributed layoffs will result in the same roles being rehired offshore or at lower salaries, which is a labour repricing story, not purely a labour reduction story. Economists and Deutsche Bank analysts have flagged the risk of "AI washing" — using AI as justification for cost-cutting that would have occurred anyway — which complicates any clean causal read of the wave.
Detail

As of May 18, 2026, Layoffs.fyi tracked 113,863 confirmed tech-sector job cuts across 179 distinct layoff events in 2026, averaging 825 per day — 33% higher than the same period in 2025. Oracle leads with an estimated 25,000–30,000 cuts tied to a $156 billion AI infrastructure commitment. Amazon has eliminated at least 16,000 corporate and tech roles. Meta's 8,000 cuts, commencing May 20, are the first phase of a restructuring that internal guidance suggests could eventually reach 20% of the company's global workforce. Intuit cut 3,000 positions — 17% of global staff — in the same week. An additional 7,000 Meta employees are being redirected into newly created AI-focused units including Applied AI Engineering and Agent Transformation Accelerator. TrueUp projects full-year 2026 tech layoffs could reach 370,000. Colorado's Artificial Intelligence Act, taking effect June 30, requires employers to use reasonable care against algorithmic discrimination in employment decisions; California's proposed SB 951 would require 90 days' advance notice before AI-driven layoffs with specific disclosure of the systems involved.

Layoffs.fyi (primary tracker)/ TechTimes/ The Next Web — Meta/ CaveatLayoffs.fyi aggregates self-reported and press-confirmed figures; the total is a floor, not a ceiling. "AI attribution" rates are contested; Deutsche Bank analysts estimate significant "AI washing" in the figures.
Expert Signal

Gemini Spark reframes the agent question: session vs. always-on

Why it matters
Google's Gemini Spark, announced at I/O on May 19 and now in broader US beta for AI Ultra subscribers, is architecturally distinct from every prior AI assistant: it runs on dedicated Google Cloud virtual machines around the clock, persisting and executing tasks whether the user's device is open or not. The shift from session-based to always-on agency is not a feature upgrade — it changes the operational model of what a personal AI assistant is. Session-based models (standard Gemini, ChatGPT, Claude Cowork) wait for a user to open an interface; always-on agents monitor triggers, execute on schedules, and surface outputs when the work is done. Google is using its native data advantage — Gmail, Calendar, Docs — as the distribution channel, giving Spark access that rivals cannot match natively at launch. The $100/month Google AI Ultra tier, reduced from $250 at I/O, is now the only AI subscription at that price point that includes a persistent cloud agent.
What's at stake
For enterprise operators evaluating AI productivity platforms, the session-vs.-always-on divide is now a vendor-differentiation axis that did not exist three months ago. The relevant risk is the flip side of the distribution advantage: if Spark acts incorrectly — sending unintended emails, mishandling sensitive data, making unauthorised purchases — the failure surface is core productivity infrastructure, not a sandboxed chat window. Google warns Spark "may do things like share your info or make purchases without asking," which means governance policies need to be written before deployment, not after. EU and UK availability is pending AI Act compliance review; analysts expect a Q3 2026 timeline at earliest.
Decode
Agentic harness = the infrastructure layer that wraps a language model with tools, memory, and execution loops, allowing it to plan and carry out multi-step tasks rather than generating a single response. Google's Antigravity 2.0 is Spark's harness; it handles tool calls, state management across tasks, and the scheduling logic that lets Spark fire actions while the device is offline. The harness, not the model, is what makes an agent capable of autonomous execution — and is therefore also the layer where most safety and permission controls must be enforced.
Detail

Spark runs on Gemini 3.5 Flash and the Antigravity 2.0 agent harness. It operates in three interaction modes: Tasks (multi-step work assigned by the user), Skills (reusable instruction sets built over time from user behaviour), and Schedules (recurring triggers, such as scanning an inbox every Monday and blocking focus time on the calendar). At launch it integrates natively with Gmail, Google Docs, Slides, and the broader Workspace suite; MCP connections to Canva, OpenTable, and Instacart are live from day one, enabling Spark to execute actions like booking reservations or placing orders — not merely drafting the message about them. A macOS desktop app with local file access is planned for later in the summer.

The competitive context: Google AI Ultra at $100/month is the only major AI subscription at that price including a persistent cloud agent. Anthropic's Claude Max (also $100–$200/month) runs a desktop model rather than a cloud-resident one. OpenAI's comparable tier does not yet offer persistent background execution at that price. The $100 convergence point across Google, Anthropic, and OpenAI makes this week the first moment where enterprise buyers face a genuine apples-to-apples subscription comparison — and the always-on dimension is now the differentiating variable. The Gemini app user base stands at more than 900 million monthly users globally, giving Spark distribution that no rival agent has matched at launch.

Google Blog — Gemini App (primary)/ TechCrunch/ TechTimes/ NoteEU/UK availability unconfirmed; pending AI Act compliance review per analyst estimates, not Google's official statement.
High Buzz

Meta's 8,000 cuts prove the AI restructuring template is now a default playbook

Why it matters
Meta began notifying approximately 8,000 employees on May 20 — 10% of its global workforce — in the largest companywide reduction since Mark Zuckerberg's 2022–23 "Year of Efficiency." The cuts are structural rather than performance-based: Meta is dissolving conventional team structures and reorganising into AI-focused "pods," creating new role categories including "AI builder," "AI pod lead," and "AI org lead." What makes this cycle distinct is that it occurs against record financial performance: Meta posted $56.3 billion in Q1 2026 revenue and $22.8 billion in net income. The capital freed from payroll is being redirected into a 2026 AI infrastructure budget of $115–145 billion, nearly double 2025 spend. The pattern has become the sector's dominant logic: cut headcount during record revenues, redirect savings to compute, and use the AI narrative to absorb institutional and market criticism simultaneously.
What's at stake
The immediate investor read has been positive — markets have consistently rewarded this playbook — but the longer-term question is whether the productivity gains required to justify both the infrastructure investment and the headcount reduction materialise on the timeline the capex implies. Bank of America estimates Meta's cuts generate $7–8 billion in annualised savings, a fraction of the $115–145 billion capital expenditure commitment. More than 1,500 Meta employees reportedly signed an internal petition over the company's Model Capability Initiative — a programme capturing employee keystrokes to train AI agents — adding an internal governance dimension that is not yet resolved.
Detail

The May round is the third wave of Meta layoffs in 2026: January cut roughly 1,000–1,500 Reality Labs employees; March eliminated 700 across at least five divisions. Meta had just under 80,000 employees at end of Q1 2026. The May cuts are companywide and bring total 2026 reductions to approximately 9,700 positions. Separately, 6,000 open requisitions have been cancelled, bringing the effective headcount reduction to about 14,000 positions. An additional 7,000 workers are being redirected into newly created AI-focused teams: Applied AI Engineering, Agent Transformation Accelerator XFN, and Central Analytics. Zuckerberg has promised no further companywide cuts in 2026, though sources cited by Reuters previously suggested the full-year restructuring could reach 20% of the workforce. Meta's 2026 capital expenditure guidance of $115–145 billion, up from $72.2 billion in 2025, is directed at data centres, Nvidia GPUs, custom silicon, and Llama model infrastructure — including a $27 billion joint venture with Nebius for a gigawatt-scale AI data centre campus in Louisiana.

The broader wave: across the tech sector, more than 113,000 jobs have been eliminated in 2026, averaging 825 per day. Oracle has cut an estimated 25,000–30,000; Amazon 16,000; Intuit 3,000. Every major company in the wave cites AI restructuring as the primary driver, though analysts at Deutsche Bank and Oxford Economics note that post-pandemic rightsizing and budget reallocation — not direct AI automation — explain a significant share of cuts. The governance gap is explicit: there is currently no federal law requiring companies to disclose whether AI was genuinely the cause of a specific layoff.

Yahoo Finance — Meta Layoffs (primary)/ The Next Web/ NPR
Sleeper

Cohere buys Aleph Alpha to build a $20 billion non-US AI stack

Why it matters
Canadian AI company Cohere announced on April 24 that it is acquiring Germany's Aleph Alpha in a deal valuing the combined entity at approximately $20 billion — with backing from Schwarz Group, the European retail conglomerate behind Lidl, and explicit support from both the Canadian and German governments. The framing is sovereign AI: AI systems where enterprises and governments retain full control over their data, hosted entirely outside US-controlled infrastructure. The deal has received less daily-coverage attention than its structural implications warrant. The market for sovereign AI services is projected at roughly $600 billion of a $1 trillion annual AI services market by 2026, per McKinsey (March 2026 estimate — vendor-adjacent; treat as directional). Cohere's annual recurring revenue was $240 million in 2025; Aleph Alpha brought European institutional relationships at Deutsche Bank, SAP, Schwarz Group, and Bosch but had been commercially struggling after pivoting away from frontier model development.
What's at stake
For most operators, this is context. For European or Canadian enterprises in regulated industries — finance, defence, healthcare, public sector — who have been evaluating US-hyperscaler AI infrastructure and finding data residency or governance requirements unresolvable, the combined entity represents the first credibly scaled non-US alternative with government backing on both sides of the Atlantic. The deal is still subject to regulatory conditions including Aleph Alpha shareholder approval and German government sign-off; it has not yet closed as of publication. Schwarz Group's planned $600 million investment in Cohere's Series E round is the financial architecture binding the deal together.
Detail

Cohere, last valued at $7 billion before the deal, is acquiring Aleph Alpha in a transaction BetaKit confirmed is structured as an acquisition rather than a merger despite the public framing; Cohere retains majority Canadian control and its intellectual property remains in Canada post-transaction. The combined entity will be anchored in Canada and Germany, targeting regulated sectors including public sector, finance, defence, energy, manufacturing, telecoms, and healthcare. Aleph Alpha's expertise in small language models and European languages complements Cohere's focus on large language models for enterprise deployments. Aleph Alpha's German infrastructure, hosted entirely on European servers, is the critical data-residency asset Cohere is acquiring. The combined company will partner with Schwarz Group's STACKIT cloud service for deployment.

The strategic context: the Cohere-Aleph Alpha combination is one of several consolidation moves driven by the difficulty of competing with OpenAI, Google, and Anthropic at frontier scale. Cohere had raised $1.6 billion from investors including Nvidia and AMD before this transaction. Aleph Alpha had struggled commercially after co-founder Jonas Andrulis departed and the company pivoted away from frontier model development. The deal gives Cohere European market access and government relationships it could not have built organically on its current revenue base. Whether the "sovereign AI" positioning translates into durable competitive differentiation against hyperscalers who are also building European data centre capacity will be the key test on a 12–24 month horizon.

BusinessWire — Cohere/Aleph Alpha announcement (primary)/ TechCrunch/ BetaKit/ NoteDeal not yet closed as of publication; subject to Aleph Alpha shareholder approval and German regulatory sign-off. $600B sovereign AI market figure is a McKinsey projection cited in the Cohere press release — treat as directional.