U.S. restores Anthropic Fable access after export-control reversal
The rollback ends a two-week disruption for Anthropic users while leaving unresolved questions about how Washington will police frontier AI models.
By Daniel Okafor · Business Editor
3 min read
The U.S. government has lifted export restrictions on Anthropic’s Fable model, allowing the company to restore access after a two-week disruption. Fortune reported that the reversal matters beyond one company because it exposed how unsettled Washington’s system for controlling advanced AI releases remains.
According to Fortune’s Jeremy Kahn, the government had imposed controls on Anthropic’s Mythos and Fable models two weeks earlier, prompting Anthropic to disable both models for all users. Officials reversed the Mythos decision on Friday evening and followed with Fable late Tuesday, Fortune reported.
Fable is designed to handle extended tasks with a high degree of autonomy, according to Fortune. Mythos has drawn attention from cybersecurity teams because it can help identify security weaknesses so defenders can fix them before attackers gain access to comparable tools, Fortune reported.
Policy questions remain
The reversals eased immediate pressure on Anthropic and its investors, but Fortune’s Kahn wrote that the episode left the U.S. with what amounts to an informal approval system for frontier AI models. He described the current approach as ad hoc, with unclear rules that appear to be set case by case by federal officials.
The Financial Times reported that U.S. officials are working with leading AI labs on voluntary cybersecurity standards. Under that approach, companies that meet the standards could have a clearer basis for expecting the government not to oppose a public model release, according to the FT.
Anthropic also said it is working with the U.S. government on a common framework for measuring the risk created when users bypass a model’s safeguards. The company said Amazon, Microsoft, Google and other members of its Glasswing partner group are involved at the start, and that it is open to participation by others.
Fortune reported that Glasswing is Anthropic’s name for a coalition of critical infrastructure companies that have been given access to Mythos. Kahn noted that OpenAI was not named among the initial participants in Anthropic’s framework effort.
Customers weigh fallback options
The short-lived restrictions may have already changed how companies view U.S.-made frontier models, according to Fortune. Kahn wrote that potential customers, especially in Europe, are now more aware that relying on American frontier systems for critical work could carry policy risk.
Fortune reported that more enterprises in the U.S. are discussing open source models as possible backups. That creates another problem for Western companies because the strongest open source models are made by Chinese AI firms, according to Fortune.
Companies can run those models on their own cloud systems to reduce data-leakage concerns, Fortune reported. Even so, Kahn wrote that businesses could face reputational risks, along with the possibility that U.S. policy may later restrict American firms from using Chinese models.
The Wall Street Journal reported that Zhipu AI’s GLM-5.2 had matched Anthropic’s Mythos in some cyber capabilities, citing a cybersecurity research firm. Fortune’s Kahn wrote that the comparison appeared narrower than a full match: GLM-5.2 could identify many of the same software flaws, while Mythos is described as able to link vulnerabilities into working exploits and use them in hacks.
Kahn wrote that safeguards are harder to enforce for open source models because screening systems can be removed and trained-in guardrails can be bypassed when users have access to model weights. He cited research showing that access to weights makes it possible to find jailbreaks against trained-in protections.
The Guardian reported that Five Eyes intelligence agencies warned recently of a near-term cyber threat from advanced AI models. The Financial Times reported that OpenAI CEO Sam Altman has renewed calls for a U.S.-led international AI governance system in which Western governments would share standards and access to powerful AI technology.
This story draws on original reporting from Fortune.