TODAY'S TOP STORIES — MAY 25, 2026
- Pope Leo XIV + Anthropic at the Vatican — Magnifica Humanitas published today; Chris Olah speaks alongside the Pope
- Gemini 3.5 Flash is GA — $1.50/$9 per 1M tokens, 4x faster than competing frontier models, now the Gemini app default
- Adobe, Canva, CapCut join Gemini — image and video editing tools accessible from inside the Gemini app
- TeamPCP GitHub attack widens — 500+ packages hit; OpenAI, Mistral AI, and the European Commission confirmed as victims
Pope Leo XIV Publishes Magnifica Humanitas — With Anthropic's Chris Olah at His Side
Pope Leo XIV today published his first encyclical, Magnifica Humanitas ("Magnificent Humanity"), at a press conference in the Vatican's Synod Hall at 11:30 a.m. local time. The document centres on "the protection of the human person in the age of artificial intelligence" — and in a first for any papal social encyclical, the Pope himself attended the launch event to present it personally.
Standing beside the Pope on stage was Christopher Olah, co-founder of Anthropic and head of the company's interpretability research team. The pairing is deliberate. Anthropic has become publicly identified as the AI company that refused to remove safeguards preventing its models from being used for lethal autonomous weapons or mass surveillance — a stance that led the Pentagon to designate it a supply chain risk earlier this year.
The encyclical was signed by Leo on May 15 — the 135th anniversary of Rerum Novarum, the foundational Catholic social document that addressed industrial-era labour rights. That deliberate date signals how Leo XIV is framing this moment: AI is this generation's Industrial Revolution, and the Church intends to be a moral stakeholder in how it develops.
Other speakers at the Vatican event include Cardinal Víctor Manuel Fernández (Dicastery for the Doctrine of the Faith), Cardinal Michael Czerny, Professor Anna Rowlands of Durham University, and Professor Leocadie Lushombo of the Jesuit School of Theology. The event and document are expected to become a reference point in every policy conversation about AI ethics for the rest of 2026.
WHY THIS MATTERS FOR AI BUILDERS
The Vatican's explicit alignment with Anthropic — the lab suing the Trump administration over military AI restrictions — turns a religious document into a significant geopolitical signal. Expect it to be cited in EU AI Act implementation debates and U.S. Senate hearings through Q3 2026.
Gemini 3.5 Flash: Generally Available at $1.50/$9 Per Million Tokens
Launched at Google I/O on May 19, Gemini 3.5 Flash is now fully generally available. As of today it is the default model in the Gemini app and Google's AI Mode in Search globally — meaning hundreds of millions of users are already running on it whether they know it or not.
Pricing via the Gemini API is $1.50 per million input tokens and $9.00 per million output tokens. Cached input runs $0.15 per million. That sits about 25% below Gemini 3.1 Pro ($2.00/$12.00) while posting higher benchmark scores for coding and agentic tasks. The API model ID is gemini-3.5-flash — no preview suffix.
Key Benchmark Numbers
| Benchmark |
Gemini 3.5 Flash |
What It Tests |
| Terminal-Bench 2.1 |
76.2% |
Agentic coding in terminal environments |
| MCP Atlas |
83.6% |
Scaled tool-use reliability |
| CharXiv Reasoning |
84.2% |
Multimodal chart understanding |
| GDPval-AA Elo |
1656 |
Real-world agentic task performance |
| Output speed |
4× faster |
vs. other frontier models (tokens/second) |
One honest limitation: 3.5 Flash is more expensive than earlier Flash-tier models. It costs 3x more per token than Gemini 3 Flash Preview and 6x more than 3.1 Flash-Lite. It is also a verbose model — Artificial Analysis found it generates roughly twice the output tokens of models at a comparable price point, which matters if you're billed per output token in high-volume workloads. Gemini 3.5 Pro, the larger sibling, is confirmed for June 2026.
PRACTICAL TAKE: WHO SHOULD SWITCH NOW
If your use case involves multi-step agent workflows, coding assistance, or tool-calling chains: 3.5 Flash is worth testing immediately. If you're doing long-document retrieval or raw knowledge-depth tasks, wait for 3.5 Pro. Free-tier access via AI Studio is available with 1,500 requests/day.
Adobe, Canva, and CapCut Add Gemini Integrations
Adobe, Canva, and CapCut have all announced Gemini integrations this week, letting users access image and video editing tools directly from inside the Gemini app. The integrations follow Google's push to position Gemini as an ambient AI layer rather than a destination chatbot.
For Canva and CapCut users specifically, this means design tasks can be initiated, described, and iterated within a Gemini conversation — the finished export then goes to the native app. Adobe's integration connects to its Firefly generative suite. No separate login or app switching is required once accounts are linked. Rollout is phased across regions through June 2026.
TeamPCP GitHub Attack: Confirmed Victims Include OpenAI, Mistral AI, and the European Commission
The TeamPCP supply chain attack — which started May 11 when attackers compromised TanStack's npm router ecosystem via the Mini Shai-Hulud worm — has now spread to at least 500 packages. GitHub confirmed the breach last week; the full victim list continues to grow.
Confirmed affected organisations include OpenAI (two employee devices, limited credential exfiltration from internal code repos; iOS, macOS and Windows code-signing certificates rotated; macOS signing cert being fully revoked June 12), Mistral AI (one developer device, facing a $25,000 Monero extortion demand), the European Commission's public website, and data contracting firm Mercor. CVE-2026-45321 carries a CVSS score of 9.6. Trend Micro has tracked at least seven confirmed TeamPCP attack waves in 2026 alone.
If you run any npm or PyPI packages in production environments and haven't audited your dependency tree against the known affected package list, do that today. The Checkmarx KICS and LiteLLM ecosystems are specifically flagged.
Also in Brief
- OpenAI + Anthropic IPO watch: Both companies are reported to be targeting public listings in 2026 — OpenAI aiming for fall, Anthropic expected to follow. SpaceX's S-1 prospectus, filed ahead of its own expected IPO, includes disclosure of its $1.25B/month compute arrangement with Anthropic.
- Anthropic revenue milestone: Reports from the week of May 23 put Anthropic at $10.9B annualised revenue and its first-ever profit quarter, up from $9B at year-end 2025.
- Google token volume: Google CEO Sundar Pichai confirmed at I/O that Google is now processing more than 3.2 quadrillion tokens per month — up from 480 trillion at I/O 2025.
- Trump AI EO delayed: Sources told Axios the executive order on AI regulation was delayed because the president "just hates regulation." He spoke with Zuckerberg, Musk, and David Sacks before the delay was confirmed.
Frequently Asked Questions
What is Magnifica Humanitas and why does it matter for AI?
Magnifica Humanitas is Pope Leo XIV's first encyclical — a major teaching document — published May 25, 2026. It centres on human dignity in the age of AI. Its significance comes from its timing (the Pope framed it as this era's equivalent of Rerum Novarum during the Industrial Revolution) and from Anthropic co-founder Chris Olah presenting it alongside the Pope, aligning the world's largest religious institution with the AI lab that has staked its identity on safety and ethics.
How does Gemini 3.5 Flash pricing compare to GPT-5.5 and Claude?
Gemini 3.5 Flash is $1.50/$9.00 per million input/output tokens. For context: GPT-5.5 API pricing sits at the premium end of the market, while Claude Sonnet 4.6 is $3/$15 per million tokens. 3.5 Flash is cheaper than both on headline pricing, though its verbose output behaviour means real-world costs on long agentic runs may be higher than the headline rate suggests.
What should developers do about the TeamPCP npm supply chain attack?
Audit your dependency tree against the confirmed affected packages (TanStack router ecosystem, Checkmarx KICS, LiteLLM, Aqua's Trivy scanner, and Bitwarden CLI are all flagged). Rotate any credentials stored in affected environments. CVE-2026-45321 (CVSS 9.6) is the primary tracking identifier. Check GitHub's own security advisory page for the current full package list.
Is Gemini 3.5 Flash better than Gemini 3.1 Pro?
For coding and agentic tasks: yes, and at a lower API price. For long-context retrieval and knowledge-depth benchmarks (MRCR v2, Humanity's Last Exam), Gemini 3.1 Pro still leads. The rule of thumb: if you're building agents or coding assistants, switch to 3.5 Flash now. If you need deep document understanding at 128k+ context, hold for 3.5 Pro in June.