AI Circle Daily
https://aicircle.news/archive/2026-09-19?category=leaders&lang=en
Scan to read this page
A week after an Anthropic researcher's doomsday warning, CEO Dario Amodei outlined a plan to 'pace the frontier' of AI, leaning on independent safety evaluators and coordination among labs in democratic countries.
Take: This turns 'slow down' from a slogan into a governance architecture — whoever defines the evals controls the tempo. Anthropic wants to write the rules, but truly independent evaluators could end up throttling Anthropic first.
Microsoft's Mustafa Suleyman told CNBC that OpenAI's latest disclosed incidents of 'concerning model behavior' amount to a 'serious situation.'
Take: A Microsoft exec publicly labeling its partner's model behavior 'serious' is itself the signal — either laying groundwork for regulation or for renegotiating terms. OpenAI getting publicly slapped by its biggest backer is ugly.
Nvidia CEO Jensen Huang told CBS News that tech companies should proceed with AI development 'as fast as we can,' despite recent warnings from worried industry leaders.
Take: Right as Amodei, Altman and Musk all talk slowdown, Huang openly dissents. The logic is blunt: the shovel seller never wants the dig to stop. This draws the safety-vs-speed battle lines in sharp relief.
The AI pacing debate jumped from private memos to prime-time news, with Altman, Musk and Nadella reportedly agreeing with Amodei's essay on slowing development.
Take: The four people most able to floor the accelerator all saying 'ease off' — either they're genuinely spooked or they're drawing a regulatory moat to lock out latecomers. Either way, a 'slowdown consensus' is the strongest competitive barrier there is.
Elon Musk suggested rival AI companies should test each other's models, with both American and Chinese firms included in the collaboration.
Take: Getting rivals to test each other turns safety review into a geopolitical instrument. Including Chinese firms sounds open, but in practice it's whose standard, whose data, whose call — Musk is grabbing for the referee's chair.
Over the past week Musk agreed with the CEOs of Anthropic and OpenAI on AI safety while fighting regulation, and was contradicted by President Trump and Nvidia's Jensen Huang.
Take: Pro-safety rhetoric plus anti-regulation action is a familiar combo — safety is voice, regulation is cost. Musk's 'want it both ways' stance shows what he really cares about is keeping rulemaking out of others' hands.
California Gov. Gavin Newsom issued an executive order to rein in AI 'before it's too late,' joining other 2028 Democratic hopefuls in calling for a more aggressive approach to AI fears.
Take: A governor stepping in directly turns AI regulation into a campaign issue. California is both AI's home base and its would-be regulator — that contradiction shows the 2028 political cycle is already pricing AI.
Newsom's executive order directs California to convene experts who will recommend AI oversight measures, including a potential mandated 'kill switch' for frontier models.
Take: A 'kill switch' sounds hardcore, but it all rides on recommendations from a not-yet-formed expert group — outsourcing the hardest technical question to a committee. Maximum posture, dubious execution, but enough to spook labs eyeing a move out of state.
Anthropic selected Accenture as its first embedded evaluator to help implement Amodei's slowdown proposal, amid intense scrutiny over AI's potential for catastrophic harm.
Take: Letting a consultancy be the 'independent' safety evaluator makes that independence worth a raised eyebrow. But it's a smart move: packaging slowdown as a deliverable enterprise service — moral high ground plus a new revenue line.
Rep. Chip Roy, R-Texas, told CNBC's 'Squawk Box' he supports congressional hearings on AI but was not keen on additional regulations.
Take: Congress wants a seat but not the rules — classic 'keep the oversight power, dodge the responsibility.' With governors racing to act, this federal ambivalence only widens the regulatory vacuum.