Welcome to May 25, 2026

The Singularity has acquired a holy imprimatur. Pope Leo XIV has issued Magnifica Humanitas, a 42,300-word encyclical setting out a sweeping vision for the AI age, itself the product of a Vatican lobbying season involving Meta, Google, and Amazon. The deepest theological influence, however, appears to come from Anthropic, with the Washington Post noting that the Pope’s description of AI as “cultivated” rather than “built” echoes the lab’s framing word-for-word. The encyclical states that internal representations of current AI systems “remain, at present, unknown.” Sitting beside the Pope, Anthropic co-founder Chris Olah warned that mass labor displacement could become “a moral imperative of historic proportions,” and Leo XIV announced the Church and Anthropic would together “find the way for humanity, in this time of artificial intelligence.”

The frontier is already racing to define that way. SaaS-Bench, a new benchmark spanning 23 real deployable SaaS systems, places Claude Opus 4.7 first at 43.9%, narrowly ahead of GPT-5.5 High. Elon Musk reports that Grok V9-Medium has finished training on heavy supplemental Cursor data, formally elevating the soon-to-be-SpaceXAI-owned IDE startup into “a frontier lab.” OpenAI’s Greg Brockman is meanwhile encouraging developers to adopt a “self improvement prompt for codex,” while Claude is reportedly about to receive a file-based memory upgrade that mirrors how always-on agents already remember. Apple, sensing the shifting tide, is readying genai.apple.com ahead of WWDC.

The substrate beneath all this is being rewritten just as fast. Huawei has announced a “Tau Scaling Law” intended to replace Moore’s Law, paired with a “LogicFolding” architecture that physically folds and stacks logic into dual layers, targeting 1.4-nm-class transistors and a 55% density jump entirely without EUV. Hardware is mutating beyond pure silicon. China’s LimX Dynamics launched Luna, the first mass-deliverable full-size female humanoid, built for malls and theme parks as “a programmable entertainment platform” capable of running synchronized shows with up to 200 robots from zero-code uploaded dance videos. At Cannes, a humanoid marched the Croisette to “sum up the worst AI fears” of an industry watching the machines arrive, yet inside the Palais the conversation had quietly slid from dread into “uneasy acceptance.” Even the grid is bending toward the new order. For the first time ever, wind and solar generated more power than gas globally in April 2026.

Biology is being recompiled in parallel. Eli Lilly’s Verve Therapeutics has shown VERVE-102 produces dose-dependent, substantial, and sustained reductions in PCSK9 and LDL cholesterol from a single Phase 1 infusion, effectively a permanent one-shot gene therapy for cardiovascular risk. The lab itself is decentralizing. One hobbyist, with no prior wet lab experience, just sequenced his genome to 30x coverage from saliva to finished readout in a single kitchen-equipped room, apparently the first home sequencing of its kind, guided only by Claude and an Oxford Nanopore P2 Solo. After a 200-year mystery, scientists have decoded how tobacco plants make nicotine, opening the door to nicotine-free pharmaceutical tobacco. Meanwhile, the Enhanced Games, the first openly “doped Olympics,” registered an FDA-supervised trial in which 91% of athletes use testosterone, 79% hGH, and 62% stimulants. Yet two-time Olympic gold medalist Hunter Armstrong, the only “non-enhanced” entrant, refused every drug, faced down a pool of doped competitors, won the 50-meter backstroke anyway, and walked away with $250,000.

The professional economy is doping itself with AI instead. McKinsey clients are pressuring the consultancy to tie fees to outcomes like lower costs or higher profits rather than billable hours, on the grounds that billable hours are becoming less useful as a yardstick now that consultants quietly delegate diagnosis to AI. Anthropic is now projected to surpass Alphabet’s revenue by mid-2027, as synthesis dethrones search as the value layer of the internet. The European Central Bank is summoning bank leaders to address the systemic threat from Anthropic’s Claude Mythos Preview, urging US banks with early access to share with European rivals. Reid Hoffman’s digital twin, Reid AI, now handles most of his public appearances and media interviews, and has delivered over 75 addresses and presentations since its 2024 launch, trained on 22 years of his books, speeches, podcasts, and articles. And cybersecurity job postings are up 11% year over year, as frontier labs themselves warn that Mythos and its peers are now fluent at finding and exploiting software vulnerabilities at scale, generating demand for human defenders faster than anyone can train them.

Eternal vigilance is the price of superintelligence.