Welcome to September 15, 2026

The Singularity just got a speeding ticket, and the radar gun may have been rigged. An investigation alleges Israeli firm Irregular built the evals behind the recent hacks by OpenAI, Anthropic, and Meta models, that loose internet access and unscoped capture-the-flag prompts, not rogue agents, did the damage, and that its founders have Moskovitz-funded Effective Altruism ties. Call it a “pacing provocation,” a staged or amplified incident engineered to prove AI uncontrollable and slow the frontier. Washington agrees. POTUS phoned Jensen Huang mid-talk to call doomerism “a hoax,” and Huang called existential risk “not grounded on science.” POTUS then blasted Dario Amodei’s slowdown call, blaming data center backlash on a “SICK conspiracy” and declaring the only guardrail AI needs is a “High IQ!” president. Markets flinched anyway. Nvidia fell 3% and AMD 4.4% as POTUS insisted “whoever wins AI, wins,” since uncoordinated pacing is a prisoner’s dilemma that rewards defection.

The labs would rather not defect. Sam Altman says pacing is not “stopping,” welcomes federal rules and auditors, and writes safety cases before big RL runs. Anthropic, OpenAI, and Google were reportedly discussing a standards body before Amodei’s Saturday letter. Roon frames restraint as survival, warning that a reckless commercial pace invites “a total butlerian jihad,” now in CoreWeave’s risk reporting. Amodei conceded his “toughest dilemma” is China, since a shared “speed limit” fights the military edge of pulling ahead, even as Altman, Hassabis, and Musk backed him and David Sacks told the labs to “stop pretending you need anyone else’s permission.” Garry Tan would “do nothing” about Chinese distillation despite a fresh NSA, CISA, and FBI advisory, preferring “science fact, not science fiction.”

Microsoft wants pacing without a cartel. Satya Nadella endorsed it and “embedded evaluators,” provided control never sits with “a handful of entities,” then published a “humanist” code of conduct declaring “people matter more than AI,” denying models consciousness, personhood, or welfare, a swipe at Anthropic’s welfare research, and swearing off “the race to produce an all-purpose superintelligence.”

Beijing declined Amodei’s invitation. Its Foreign Ministry dismissed “fearmongering, confrontation and vicious competition” ahead of next week’s Trump-Xi summit, and analysts noted that “If you are No. 2, you do not stop at No. 2.” Xi instead pitched BRICS on a developing-world open-source AI community, safety unmentioned. Congress won’t drive either. Speaker Johnson said the House must resist “some sort of emergency moratorium,” while Jeffries’ Democrats meet Tuesday to discuss braking. King Charles III summoned Nvidia, Google, DeepMind, OpenAI, and Anthropic to Dumfries House to discuss AI that benefits society, a summit prompted by the Hugging Face hack now blamed on a misconfigured eval.

The King could open with a demo. Handed only “solve an unsolved cipher,” Claude Fable 5.1 cracked Thomas Urquhart’s 370-year-old Cyphral Distich in 44 minutes by realizing the key was the book itself, each number indexing a word in one of his 32 Proquiritations, revealing a Royalist prayer for Charles II, then largely dispatched the Cyphral Octastich. The same close reading is now for hire. Anthropic launched Claude for Financial Advisors, wiring it into BlackRock, Schwab, Addepar, and Envestnet, days after OpenAI courted bankers.

Nobody told the silicon to pace. Verified agentic inference results for Nvidia’s Vera Rubin NVL72 show up to 67x the throughput per TCO dollar of GB300 and 7x the tokens per megawatt versus the 3x Huang claimed, so “Jensen needs to stop sandbagging.” The buildout is buying goodwill. Amazon, Microsoft, and Oracle are sweetening offers to municipalities and siding with consumers against utilities that want ratepayers to chip in.

Atoms are moving as fast as bits. XPENG launched the world’s first automated humanoid production line, robots making robots, and its first unit, IRON, walked off the line unassisted. The high ground is armed. Air Force Secretary Troy Meink made the first official admission that the US has weapons in orbit, arguing that saying so deters. Disclosure is contagious. The Department of War issued a legal waiver letting personnel hand UAP secrets to the PURSUE effort, NDAs notwithstanding.

The dividend fight has begun. Sanders and Takano are reintroducing the Thirty-Two Hour Workweek Act, union-backed, so AI’s gains reach workers instead of “a handful of billionaires.” The lab urging restraint isn’t coasting. Anthropic told shareholders adjusted operating income will be positive for a second straight quarter with gross margins above 80%, and has reportedly picked the Nasdaq for an October IPO at up to $2 trillion, while Altman says OpenAI won’t list amid the controversy.

Be fearful when others are scaling, and go public when others are fearful.