White House ordered Anthropic to pull Fable — and SoftBank is funding OpenAI with junk
Politico's tick-tock of the 19-day standoff is out, and the same week SoftBank went to the junk-bond market to keep its OpenAI stake growing. Here is what both say about how the AI build-out is actually governed and financed.
Politico has reconstructed the 19 days the White House spent trying to get Anthropic to pull Fable 5 — and it ends with the government doing it for them. The magazine piece, published Sunday, walks through the spring sequence: Amazon chief executive Andy Jassy raised jailbreak concerns about the model with the administration, which Fortune reported was the phone call that started the crackdown; Commerce then hit Fable 5 and Mythos 5 with export controls, taking both offline. Politico reports that aides spent hours trying to reach Dario Amodei and were told he was at a wellness retreat, before delivering what it describes as a blunt set of directions: take Fable down and work with the government to get it fixed. Anthropic complied, the controls were lifted once the company addressed the jailbreak, and Fable 5.1 shipped on schedule in September.
What matters is not the history — it is the precedent. This is the first documented case of Washington ordering a deployed frontier model offline, and the mechanism used was export control, a tool designed for shipping crates rather than API endpoints. Anthropic was not fined or sued; its product was simply switched off until it satisfied the government, which makes capability evaluation a licensing regime in all but name. The uncomfortable part is that the same administration spent last week calling AI safety fears a "hoax" and announcing an "AI Force" while, per WIRED, officials still want Anthropic to block all jailbreaks before it considers future releases — a bar security researchers say may not be technically reachable. The politics changed; the switch did not go away.
We covered the legal fallout of that pressure campaign in August — Judge strikes down Pentagon's blacklist of Anthropic.
SoftBank is going back to the bond market for OpenAI, and this time it is junk. Reports citing the deal's term sheet say the group plans to issue $10 billion in dollar-denominated notes across three tenors plus €1 billion (about $1.1 billion) in euro notes across two maturities, with proceeds partly funding a follow-on OpenAI investment expected to close next month. The reports put SoftBank's total OpenAI commitment near $65 billion, and note it recently cleared the remaining $25.9 billion balance on the $40 billion bridge loan it took in March. Bloomberg describes the sale as potentially one of the largest junk bond offerings ever.
The shift is the story. Two years of frontier-AI money came in as equity, where the downside sits with venture funds and sovereign investors who signed up for it. This round is high-yield debt, which means the risk of a company that has never posted a profit is being spread to bond buyers hunting yield — and the collateral is ultimately a mark on a private stake nobody can price until OpenAI lists.
Tencent's Marvis stopped being an assistant and became an "AI butler" — four of them, in fact. The company's OS-level PC agent shipped a "butler team" update, per 智东西: computer, file, browser and software agents now run side by side, with five more (games, finance, life services) promised and the butler framework opening to outside developers — campus, visa and pet butlers were the examples given. The computer butler diagnoses CPU, memory, disk and driver problems in natural language and can set up a local model deployment end to end; the file butler builds a shared semantic knowledge base over documents and images; the browser butler runs scheduled web tasks and hands the results back; the phone app can take over the machine remotely when a task stalls.
The roadmap is the tell. Tencent says Kylin and UOS builds — China's state-backed operating systems — arrive in October for government and enterprise PCs, with the software butler limited to automating apps covered by a Kylin licensing agreement, and it is working with domestic OEMs to put a Marvis entry point and control panel inside the OS itself. That is an assistant trying to become the device layer rather than a window on top of it.
What to watch: whether the White House's jailbreak bar becomes a written standard applied to every lab, or stays an unwritten condition Anthropic happens to be living under.
Should a government be able to switch off a deployed model — and if so, what should the trigger be? Tell us in the comments.
