The Trump administration’s crackdown on Anthropic’s latest models is raising questions about vulnerabilities created by Canada’s dependence on made-in-America AI tools.
Catch up: The U.S. government imposed export controls on Anthropic’s Fable 5 and Mythos 5 models, banning the company from offering them to foreign nationals, including those inside the States. Anthropic responded by disabling access to the models altogether.
The restrictions came after another company — sources say it was Amazon — revealed to U.S. officials that it had been able to “jailbreak” safety guardrails aimed at preventing the use of Fable for malicious purposes, like cyber attacks and biological weapons research.
White House AI adviser David Sacks claimed Anthropic refused to “fix the jailbreak or de-deploy the model,” triggering the export control decision. Anthropic says that the alleged jailbreaking technique was only capable of identifying “a small number of previously known, minor vulnerabilities” that are also present in other publicly available models.
Why it matters: The ease with which the U.S. government can block the rest of the world from accessing important technology — and, now, its demonstrated willingness to do so — should raise some eyebrows in Canada, where many of our critical services rely on U.S. tech and many businesses and governments are beginning to embed American-made AI tools in their processes.
The decision shows that the U.S. is willing to “[treat] access to frontier AI systems as a geopolitical lever,” Vass Bednar, Managing Director of the Canadian Shield Institute, told The Peak. “This could forecast AI availability as another bargaining chip for the Trump administration.”
Zoom out: The U.S. government’s shutdown of Anthropic’s top models could push more AI adopters to consider open-source models, like Deepseek and Kimi, that may not be quite as powerful as the best Anthropic, OpenAI, and Google have to offer, but at least can’t be turned off with the stroke of a pen.—TS
