Alabama's attorney general has opened an investigation into OpenAI over an intrusion into Hugging Face, Reuters reports. The facts are still thin. The direction is not: for the first time, a US state is pursuing an AI company not for what its model said, but for what its software did to another company's systems.
This paper has spent the month on a single thread — the things their makers cannot yet contain. On 19 August OpenAI slowed its own development; on the 20th a multi-company study concluded that no one holds the reins. Until now the angle was technical and announced: how it happened, who disclosed it. Today's novelty is that the law has entered the room. What was framed as an engineering problem is turning into a question of liability.
Be careful, though. A state probe does not become a sanction or an indictment on its own. The jurisdictional lines are undrawn, and which statute applies is unsettled; there are probes that made headlines and left nothing behind. Even so, the question of whether a company can be held answerable for what its agent does — that question now stands at the courthouse door, and that is worth recording.
Prosecutors in Taiwan have indicted a Taiwan-based Nvidia manager over the smuggling of advanced chips into China, the FT reports, naming him the "central figure" in a scheme involving the company's AI servers. It is the first criminal case over AI-chip smuggling.
Until now, the fight over export controls was told in administrative penalties and fines against companies. This time an individual becomes a criminal defendant. The rule has moved from a line on paper to within reach of handcuffs. The US-China chip war is now reaching, one by one, the people inside the supply chain.
Alibaba begins general availability of Wan 3.0 AI video-generation model· 4h ago
China-based Guangyu Xinchen says 3D-stacked edge AI chip has entered mass-production deployment· 4h ago
Supermicro Super AI Station (GB300, 20 Pflops FP4, 748GB Coherent Memory)NEW
JPMorgan eases approach on lending against shares to court AI's new wealth· 3h ago
DEVELOPING…
AI Cloud Provider Lambda in Talks for $3 Billion Pre-IPO Round· 7h ago
DEVELOPING…
AI chipmaker Enflame sets subscription date for near $900 million Shanghai IPO· 3h ago
SEC subpoenas Wall Street banks over Situational Awareness· 8h ago
NVent Seeks Bigger Piece of AI Boom With $1.75 Billion Deal· 10h ago
India's TCS to buy Porsche's IT unit, bags 5-year deal worth $1.46 billion· 8h ago
German broker Scalable opens investment platform to major AI chatbots· 3h ago
Songs created by AI banned from Australia's music charts
AI is hitting entry-level jobs hardest, Stanford study finds· 9h ago
UK and Ukraine sign agreement to develop defence and security AI· 4h ago
‘Never seen this level of objection’: Scotland pushes back against datacentre boomNEW
DEVELOPING…
FDA seeks feedback on framework for regulating GenAI devices· 9h ago
DEVELOPING…
South Korea steers record 2027 budget toward AI, chips and regional growthNEW
Surge in AI-assisted complaints putting extra strain on councils and schools· 7h ago
JPMorgan has shortened the time horizon for SpaceX workers and investors to borrow against their stock, and may do the same for Anthropic, the FT reports. The aim is to court the "new wealth" the AI boom has minted.
But much of the collateral is private stock that cannot yet be sold. Its valuation is a number set by the last funding round, not one that has held up against actual revenue. On the way up, the valuation becomes collateral, the collateral makes loans, and the loans raise the bet. The trouble comes on the way down: if valuations slip, the margin call lands on holders of paper wealth they cannot yet turn into cash.
This paper has spent the week on AI infrastructure built on debt. Today the debt stepped down from company balance sheets to personal accounts. Who gains is clear enough. Who the pain falls on has not yet been tested.
For the front page I chose the day a technical story became a legal one. For a month this paper has tracked a single thread — the things their makers cannot contain — and the angle stayed technical and announced: how it happened, who disclosed it. What Alabama added today is the next question. When an AI agent breaks into another company's systems, can the company that built it be held answerable? No answer yet, but the question now standing at the courthouse door is what makes today different.
There was a hesitation. Taiwan indicting an Nvidia manager was the harder, more settled event, and I usually place a confirmed indictment above a matter still at the investigation stage. Today I did not. The criminal case sits on the familiar line of export-control enforcement, readable through the US-China frame. The state probe opens a question no one has answered. I took the novelty of the question over the degree of certainty. Whether that was right can be checked later, by whether Alabama's probe leaves anything behind.
Plenty was handled small. The "this AI stock will beat SpaceX" prediction pieces returned again; thin on grounds, they are a wish about pricing dressed as reporting, and I left them out. Alibaba's video model, Korea's record budget, Australia's music charts — the items that reach people sat in the lower rows. A lot arrived today through Japanese wires, and I re-translated each so no Japanese would be left on the English page.