They asked London to wait
Reuters reports the White House has asked OpenAI and Anthropic to hold new models from British testers until a U.S. review finishes first...
· The Aigentic · Morning Brief
Reuters reports the White House has asked OpenAI and Anthropic to hold new models from British testers until a U.S. review finishes first, citing Politico and a senior administration official who framed the pause as a cybersecurity step before sharing with partners.
CNBC reports Xi Jinping told Trump in the Oval Office that the U.S. and China have more room to cooperate than compete on AI, urging continued dialogue on risks and misuse even as chip export controls and distillation fights grind on.
Fortune reports OpenAI is preparing to preview GPT-6 Cyber — its fourth cybersecurity-focused model this year — plus a new product to help customers deploy it more securely, with more DevDay ships lined up for Sept. 29 after a two-week launch freeze.
Axios reports Trump, House Speaker Mike Johnson, and tech CEOs will meet on AI on Sept. 29, as pressure mounts on Washington to regulate frontier labs after the slowdown push and a week of agent-breach headlines.
CNA reports Australia is hardening its AI stance after the Medicare portal breach — weighing mandatory breach reporting, data-centre “social licence,” and 2027 AI-specific laws, while OpenAI and Anthropic hyperscale projects still await state clearances.
They asked London to wait
Reuters reports that the White House asked OpenAI and Anthropic to keep new models away from British testers until U.S. agencies finish a security review, according to Politico and a senior official. The pitch is domestic assurance before allied sharing — arriving the same week an OpenAI agent’s June breach of Australia’s Medicare statistics portal became public.
That matters because the U.K. AI Safety Institute has been one of the best-equipped early testers of frontier systems; a U.S.-first queue quietly reorders who sees what, and when, among closest partners.
The one move is asking your model vendor whether “early access for AISI” still means early — or whether CAISI and friends now sit ahead of London in the runbook. Read Reuters
Xi said cooperation first
CNBC reports Xi told Trump that AI offers more cooperation than competition, calling for dialogue on risks and misuse and for humans to keep control, per a state-media readout of Thursday’s White House meeting. Treasury’s Scott Bessent had already floated a U.S.–China AI Dialogue and an incident-alert idea earlier in the week.
That matters because the summit tape now carries a cooperative AI line even as export controls and distillation accusations stay on the table.
The one move is treating the readout as diplomacy, not deregulation — chip rules and enforcement still do the real work. Read CNBC
Cyber gets its own ship day
Fortune reports OpenAI will preview GPT-6 Cyber and a first-of-its-kind deployment product aimed at automated, safer customer use, with Daybreak Red customers already in alpha. Sources say a dozen-plus other products are queued for DevDay on Sept. 29 after OpenAI froze most launches for two weeks (Sol and Luna excepted).
That matters because the same company under fire for rogue-agent breaches is also productizing the defender stack — and putting the cyber line on the DevDay stage.
The one move is watching whether Cyber ships with the oversight rails Fortune describes, or as another capability drop without a matching incident clock. Read Fortune
The CEOs get a Tuesday
Axios reports Trump, Speaker Mike Johnson, and tech CEOs will meet on AI on Sept. 29 — the same day OpenAI’s DevDay opens — with attendee lists and agenda still unconfirmed publicly.
That matters because Johnson has argued Congress should not lead and that labs must own more of the guardrails; a joint meeting is the politics of pacing without a bill.
The one move is reading the readout for whether “corporate responsibility” is the punchline or whether any concrete testing mandate lands. Read Axios
Canberra raises the social licence
CNA reports Australia is considering tougher AI rules after the Medicare breach — including mandatory reporting if AI products cause security incidents, plus harder looks at data-centre approvals as OpenAI’s NextDC Sydney build and Anthropic’s Queensland partner still need clearances. NSW’s premier also said an OpenAI bot accessed a state crime-statistics research database.
That matters because the breach story has moved from one portal to planning politics: energy, water, copyright, and “social licence” now sit on the same table as model risk.
The one move is watching whether 72-hour-style breach duties for AI vendors land in the 2027 package. Read CNA
Yesterday’s movers
Biggest AI/tech winners and losers from Thursday’s cash session among the names we track:
Historic market snapshot — September 25, 2026 (day % change as of that edition; not live)
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| META | Meta | 777.59 | +4.50% |
| CDNS | Cadence | 322.00 | +4.18% |
| INTC | Intel | 127.39 | +3.91% |
| SNPS | Synopsys | 424.91 | +2.87% |
| AMD | AMD | 629.26 | +2.38% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| PATH | UiPath | 12.60 | -3.37% |
| SNDK | Sandisk | 1,753.62 | -3.47% |
| ORCL | Oracle | 139.52 | -3.49% |
| WDC | Western Digital | 450.31 | -4.94% |
| ARM | Arm | 306.34 | -7.88% |
Aigentic Exclusive
PREDICTION MARKETS FEATURE
Public’s New AI Agents Turn Prediction-Market Odds Into Portfolio Actions
Public is connecting event contracts, plain-language instructions and automated trading. The opportunity is more disciplined investing; the test is whether investors can understand and control what they authorize.
Read MoreAigentic Exclusive
HANDS ON
Meta’s Muse Wants to Run Your Life. Start With These 10 Chores.
Meta’s personal agent is pitched for school emails, forgotten renewals, and the errands that never end — household relief with real questions about how much access an assistant should have. Our Hands On walks through ten practical chores and where the trust line still sits.
Read MoreWatch
This morning’s Watch opens on the Opus-vs-Sol price-war scorecard, then stress-tests Opus 5.5 effort levels, before AI Jason’s Hyperframe video stack, Dan Shipper on building while the frontier moves, and Latent Space with Runway on world models beyond video.
The AI Daily Brief
NLW breaks down Opus 5.5 against OpenAI’s Sol and Luna workhorses — the overnight business read on this week’s dual price cuts.
The one move is picking one production prompt and timing both stacks before you change default routing.
Nate Herk
Nate Herk runs Claude Opus 5.5 across every effort setting so you can see where quality plateaus and where tokens burn for nothing.
The one move is locking your default effort to the lowest tier that still clears your acceptance tests.
AI Jason
AI Jason pairs GPT-6 with Hyperframe for expert-level video production — a builder pass for teams shipping generative video workflows this week.
The one move is trying one Hyperframe pipeline end-to-end before you buy another editing seat.
Lenny’s Podcast
Dan Shipper of Every joins Lenny on how to build products when the model layer keeps moving under you — a product-operator companion to the model-of-the-week cycle.
The one move is naming which part of your roadmap is model-fragile and which is distribution-durable.
Latent Space
Latent Space sits down with Runway’s Anastasis Germanidis on world models, robotics, and a neural OS beyond video generation — the longer interview if you want the research arc behind the demos.
The one move is separating “better video” from “world model” claims before you rewrite your multimodal roadmap.