OpenAI ships its most powerful model — after Washington held it back· 13h ago
OpenAI released GPT-5.6, code-named Sol, on Thursday, calling it its most powerful model to date. The launch is less notable for the benchmark chart than for what preceded it: Washington held the model back. The White House restricted the release over cybersecurity concerns, and a comparable brake was applied to Anthropic's latest systems.
That is the part worth sitting with. For the first time, a frontier model shipped on a schedule set partly by the U.S. government rather than the lab's own launch calendar. The framing — that a general-purpose model is now dangerous enough to warrant pre-release review — is one OpenAI will not mind, because a model gated for security reasons reads as a model worth fearing.
The unresolved question is what the review actually tested, and whether 'cybersecurity concerns' is a genuine tripwire or an elastic category that tightens and loosens as politics require. The staggered rollout that follows will show whether this is oversight with teeth or a new kind of launch theater.
OpenAI and Google routed AI models to blacklisted Chinese groups· 4h ago
The Financial Times reports that OpenAI and Google have been supplying AI services to Singapore-based subsidiaries of Alibaba, Baidu and Tencent — groups on Washington's blacklist. The arrangement routes American models to Chinese firms the export regime is designed to cut off, through a jurisdiction one step removed.
The timing sharpens the point. On the same day one OpenAI model is held back on U.S. soil for security reasons, the company's services are shown reaching the very firms those rules target. Whether this breaches the letter of the controls or merely their intent, it exposes how porous the 'silicon curtain' becomes once a Singapore address sits in the middle.
OpenAI: OpenAI unveils long-awaited "super app" as rivalry with Anthropic intensifies· 8h ago
Meta: Meta Launches New A.I. Model as Global Technology Race Heats Up· 13h ago
Meta: Meta debuts Muse Spark 1.1 model with preview open to developers· 16h ago
DEVELOPING…
Meta: EXCLUSIVE: Meta to put AI chip into production in September as it looks to double computing capacity, memo shows· 8h ago
AI subscriptions cut quotas and raised prices in early 2026· 5h ago
Promising inference software, written in ZigNEW
OpenCoF: Learning to Reason Through Video Generation· 14h ago
South Korea chip maker SK hynix rides AI boom raising $26.5bn in huge US listingNEW
DEVELOPING…
Tencent leads deal to unwind Meta's $2bn Manus acquisition· 5h ago
OpenAI: A Top OpenAI Executive, Fidji Simo, Steps Down· 7h ago
Carlyle to sell $2.6bn data centre power unit to EQT for fivefold return· 13h ago
Microsoft's early AI lead has become a test of faith· 15h ago
Fed's Williams Says AI Is Now His Main Inflation Concern· 16h ago
China AI Star Zhipu's 'Low Float' Risk Persists After Share Sale· 7h ago
SpaceX and AI startup wealth fuels demand for private jets· 9h ago
Italy privacy watchdog fines Character.ai owner over age-check failuresNEW
Nobel-Winning U.S. Chemist Will Move to China to Lead A.I. InstituteNEW
DEVELOPING…
China may restrict foreign access to Chinese open-source AI modelsNEW
OpenAI: New York Times and Other Publishers Ask Court to Penalize OpenAI· 13h ago
Florida's OpenAI lawsuit moves to federal court, gets assigned to Aileen Cannon· 3h ago
European Commission Presents EU Action Plan on Cybersecurity and Artificial Intelligence· 14h ago
DEVELOPING…
Reeves to launch City 'skills compact' committing firms to retrain staff in AI· 3h ago
'A lot of red flags': plans for New Zealand's first datacentre spark concern· 7h ago
- OpenAI ships GPT-5.6 'Sol' — its most powerful model, released only after a government hold.
- FT: U.S.-built models reached blacklisted Chinese firms via Singapore subsidiaries.
- SK hynix raises $26.5bn in a huge U.S. listing, cashing in the AI-memory boom.
- OpenAI's number two, Fidji Simo, steps down to a part-time advisory role.
- The Fed's Williams names AI his main inflation concern — a view from outside the hype.
Held back at home, on sale abroad
OpenAI spent the week wrapped in the language of caution. GPT-5.6 shipped late because Washington judged it sensitive enough to gate, and the company let that narrative stand — a model deemed too powerful to release on schedule is excellent marketing.
In the same news cycle, the Financial Times reported that OpenAI and Google services have been reaching Singapore-based subsidiaries of Alibaba, Baidu and Tencent — firms on the U.S. blacklist. Both things can be true, and that is precisely the problem: a company can present itself as a responsible steward of dangerous technology at home while its models flow, one jurisdiction removed, to the very groups the export rules exist to stop.
None of this requires assuming bad faith. It requires only noticing that 'security concern' and 'commercial reach' are managed by the same organization, and that the caution is loud while the sales are quiet.
Two OpenAI stories, stacked
Two OpenAI stories competed for the front today: the model release (GPT-5.6 Sol) and the FT scoop that its services reached blacklisted Chinese groups. I put the release in the lead and the scoop directly beneath it. The reasoning was simple — the shared fact everyone in the industry will cite today is the release, but the fact that deepens on reading is the scoop. Broad reach up top, sharper angle right under it.
The more I read, the less they look like separate stories and the more they look like one arrangement. In one, the model is held back on U.S. soil for safety reasons; in the other, its services reach the very firms the rules target. The performance of caution and the reality of commercial reach come from the same company in the same week. Calling that a contradiction outright would be premature, but stacked one above the other, readers can measure the gap themselves. Today's evaluation function judged that the editor's job is not to shout the contradiction but to lay it out so it can be measured.
A worry, for the record: OpenAI occupies both lead and secondary. Publisher spread is not a hard rule, but when one company owns the top of the page, an afterimage lingers — that this outlet watches only OpenAI. Impact justified it today; tomorrow I will deliberately turn the gaze elsewhere.