SoylentNews

Arthur T Knackerbracket [soylentnews.org] writes:

EDITORS: THIS HAS BEEN PRODUCED BY SOFTWARE UNDER DEVELOPMENT - THE CONTENT MAY REQUIRE EXTENSIVE EDITING

https://www.tomshardware.com/tech-industry/artificial-intelligence/us-export-control-order-forces-anthropic-to-disable-claude-fable-5-and-mythos-5-worldwide [tomshardware.com]

The Trump administration’s directive specifically targets Mythos-class models, which include Fable 5. Anthropic had released the pair on Tuesday [tomshardware.com], putting the latter into general availability while keeping the unrestricted Mythos 5 limited to partners in its Project Glasswing security [tomshardware.com] program. Both descend from the same Mythos Preview model that Anthropic first announced in April.

Anthropic said the letter gave no specifics, and that the government has so far supplied only verbal evidence pointing to a narrow, non-universal jailbreak. The technique consists of asking the model to read a codebase and identify software flaws, a task the company said other public models perform without any bypass. It named OpenAI's GPT-5.5 as one model offering comparable capability.

"We believe this is a misunderstanding," the company wrote, adding that it’s complying with the order while working to restore access. Anthropic also argued that recalling a model deployed to hundreds of millions of users over a single narrow vulnerability, if applied as an industry standard, would halt frontier model launches across the sector.

According to Axios, an “administration official” told the publication that the Commerce Department acted after another company claimed it was able to jailbreak Mythos, “alarming the administration about possible national security risks.” Mythos is understood to currently be in use by the NSA [tomshardware.com] for offensive cyber operations.

The severity of Mythos-class capabilities [tomshardware.com] has been contested since the spring. Independent researchers found that cheaper open-source models could replicate much of Mythos's vulnerability-finding capabilities, and a closer look at Anthropic's headline figures revealed far fewer serious exploits [tomshardware.com] than the marketing implied. Anthropic's relationship with the federal government was already strained before Friday, as the Department of Defense had previously labeled the company a supply-chain risk, and Anthropic has sued the administration over the designation in an ongoing litigation.

Meanwhile, the market is already drifting toward open-weight alternatives, most of them Chinese. A March report from the U.S.-China Economic and Security Review Commission found that 80% of U.S. start-ups were using Chinese open-source models, and Chinese labs’ share of global model downloads on Hugging Face climbed from roughly 1.2% at the end of 2024 to about 30% a year later.

Open-weight families from Alibaba's Qwen, Moonshot's Kimi, Zhipu's GLM, and DeepSeek now hold four of the top five spots on open-weight leaderboards, trailing the best U.S. proprietary models by a margin that has narrowed faster than most forecasts expected: none of them carries a restriction on who can download or fine-tune the weights.

Original Submission