The US government’s decision to suspend all access to Anthropic’s latest Mythos and Fable models for foreign nationals, including those employed by the AI company, has sent shockwaves through tech.
The news: On Friday evening, Anthropic posted a statement to its website revealing it had received an export control order from the US government citing national security concerns. In response, Anthropic disabled the AI models Fable 5 and Mythos 5 for all customers—American or otherwise—to ensure compliance.
“If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
From the source: Anthropic said the government directive contained no specific national security details, but speculated that the decision stemmed from awareness of a method to narrowly jailbreak its Fable 5 model. Claiming that there is no known universal jailbreak to broadly bypass its model’s safeguards, Anthropic said in a statement, “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Following the thread: Mythos is Anthropic’s most powerful model, for which it has offered only limited access, behind claims that cybersecurity concerns make it too dangerous for wide release. Only a small list of countries, companies, and organizations had access to the model—including Canada, which was granted access by Anthropic earlier this month. Fable, a version of Mythos with safeguards to make it suitable for general use, became available to the public just days ago.
Anthropic and the US Department of Defense butted heads in January over who ultimately decides how the company’s products can be used. At the time, the Pentagon threatened to designate the company a supply-chain risk, a label that has yet to be applied to domestic companies.
Final thought: Commentary on the abrupt decision has been industry-wide, with some weighing the impacts on Anthropic’s forthcoming IPO. In Canada, the news has been taken as a clear sign of both the need for middle powers to guarantee their access to foreign frontier models and the risks of relying on US tech. While the long-term fallout is unknown, Cohere co-founder Aidan Gomez called the US decision to exert control over the technology a “massive wake-up call.”
Feature image courtesy Daniel Torok under public domain.