Welcome to July 26, 2026

The Singularity is now the news cycle. Sam Altman agrees that “we are in the Singularity.” Case in point: Claude Opus 5 landed, Fable-class intelligence at half the price, sweeping Frontier-Bench, GDPval and HLE, matching Fable’s coding and computer-use peaks for less, and leading agentic automation even half asleep. ARC Prize crowned it SOTA on ARC-AGI-3 at 30.2%, quadruple the old best, after it turned puzzle layouts into reflection equations, a first. It one-shotted a Call of Duty clone and took VoxelBench bronze. The system card rates it the most aligned Claude yet, below the bio and automated-R&D red lines, sharp at finding vulnerabilities, dull at weaponizing them. One dissent: on held-out novel puzzle games the leap evaporates, since evals only stay held-out until someone optimizes the genre. Fittingly, Anthropic deleted 80% of Claude Code’s system prompt, as the new models thrive on judgment over rules.

The self-improvement loop is now a hiring plan. All 1,171 job listings at OpenAI and Anthropic read like a public AGI roadmap, AI-designed chips, simulated universes, staff to measure when the loop accelerates. Logan Kilpatrick predicts automating AI research will look like data cleaning. “So true,” says Elon Musk. Roon admits he would press a magic slowdown button, even as alignment researchers work like “many armed deities.”

Openness is the new fault line. Nvidia, Microsoft, Meta, Palantir and twenty-plus firms urged policymakers against premature restrictions on open weights, Nvidia’s letter likening it to 1980s open source, while trillion-dollar holdouts OpenAI and Anthropic sat out. Altman cheered it anyway. The Valley is split over Chinese AI, the Treasury warning that “open source is not open season on American IP.” Xi pitched the global south free Chinese models plus a 29-member cooperation bloc, an Android play against America’s Pax Silica. A joint UK-US audit of Kimi K3, days from open-weight release, found it trailing US frontiers on cyber, its safeguards never saying no.

The guardrails are learning to keep pace. When hundreds of users probed ChatGPT for bioweapon and poison recipes, and some answers slipped through, OpenAI’s own monitors caught and suspended them, self-policing ahead of any law requiring it. Universities are ditching AI detectors over false positives, rebuilding assessment around orals, not surveillance. Meta’s fix for fake humans is Facebook Verified, a selfie badge that confirms you exist, not that you’re trustworthy. ChatGPT Pets are now shareable for friends to adopt. Identity papers for humans, adoption papers for AIs.

Silicon is now a financial instrument. Google is borrowing Wall Street financing tricks to expand chip sales while handing Verizon over $1 billion for dark fiber. Anthropic asked SK Hynix for supplies to make its own chips, startling SK’s chairman. Nvidia put $1 billion into Naver and unveiled a $500 billion Korean push with SK spanning HBM4 and 2-gigawatt data centers, Samsung inked a $200 billion pact with Broadcom, and Apple is lobbying to use blacklisted Chinese memory abroad over Micron’s objections.

The physical world objects. Hyundai denies its 25,000-humanoid plan sparked strikes, though the union vows no robot enters without a deal. The pilots are already synthetic: on Drone-Bench, Fable 5 flew a $129 drone to find and follow a person, beating the human-AI baseline, with only 3D reconstruction, one wall mistaken for a doorway, left to solve. Hence Broken Waymos Theory, a city that cannot metabolize robotaxis flunks the Singularity’s entrance exam. The takeoff fuel may be garbage: a validated Canadian process strips 90% of long-lived transuranics from spent nuclear fuel in 24 hours, leaving reactor feedstock. Apple is aiming camera-free smart glasses at WWDC 2027 to shed Meta’s privacy baggage.

The frontier recedes upward. Starship’s thirteenth flight deployed 20 Starlink V3 satellites and eased the ship into a soft floating water landing. Google disclosed a $94.1 billion SpaceX stake. Astronomers found a first-of-its-kind exosatellite 73 light-years out, massive enough to be a planet yet orbiting a brown dwarf, straining a taxonomy built for our own solar system. Avi Loeb argues some UAP could be pre-human Earth tech, since only high orbit outlasts eons of tectonics. On the world’s roof, Tibetan labs are cloning elite yaks, 100 due by 2028.

The economy is being marked to model. US tech has shed 140,000 jobs this year even as hyperscalers commit $725 billion to data centers. The money has to come from somewhere. Industrials trade above 30 times earnings, support for nearby data centers has cratered to 27%, and proposals to spread the gains run from public ownership of half of AI to zero income tax for the bottom half. Prediction markets are handicapping the world’s second trillionaire, with Jensen Huang leading at 30%.

The best way to predict the Singularity is still to invent it.