Canada PM Warns of AI Risk After US Orders Anthropic Models Offline

In brief

Canada’s Prime Minister Mark Carney warned that a U.S. order on Anthropic shows the risk of leaning on only a few AI makers.

Washington forced Anthropic to pull two frontier models offline for every user worldwide.

Decentralized AI removes the "kill switch," but still relies on a few chipmakers, observers told Decrypt.

Canadian Prime Minister Mark Carney warned Sunday that U.S. restrictions on Anthropic's most advanced AI models expose the risk of leaning on a handful of American providers, after a government order pulled two frontier systems offline for users worldwide.

"The situation we're in collectively right now with Mythos and Fable is something that can happen with overreliance on certain models," Carney said, speaking in Ireland ahead of the G7 summit in France, the Associated Press reported.

Carney said nobody has "done anything wrong in the situation," but warned the mistake would be to "just accept this, don't take the lesson, don't build out and diversify."

“It is never a good idea to have one option,” the prime minister added.

Carney’s remarks follow a directive on Friday ordering Anthropic to cut access to its Fable 5 and Mythos 5 models for any foreign national inside or outside the United States on national security grounds.

Anthropic promptly complied, disabling both systems for all customers. It disputed the basis, however, arguing the cited jailbreak is already replicable on public models like OpenAI's GPT-5.5.

Commerce Secretary Howard Lutnick reportedly sent the letter to Anthropic CEO Dario Amodei, per Axios. It was reportedly driven in part by suspicion that a China-linked group had accessed Mythos, Semafor reported.

The disputes come as Anthropic nears a $1 trillion valuation following annualized revenue of over $47 billion.

Decrypt has reached out to Anthropic for comment and will update this article should they respond.

Risks and costs

The prime minister's concerns cast centralized AI as a single point of failure, where one company's compliance with a government order can cut off users everywhere at once.

Centralized AI keeps a model and its controls inside one company. Decentralized AI spreads those functions across independent operators coordinated by a blockchain.

Projects tied to decentralized AI rallied following the ban on Anthropic's models, with the sector's market cap at $24.3 billion, up 6% on the day and 12% over the week, per data from CoinGecko.

Smaller compute and data networks led the way, with ChainOpera AI, io.net, Grass, and NOVA among those climbing more than 30% over the past week, while NEAR Protocol and Bittensor, two of the sector's largest tokens by market cap, rose 15.9% and 27.9% on the week respectively.

The U.S. government's move on Anthropic points to "a risk that is largely unique to centralized AI," Dan Dadybayo, strategy lead at Horizontal Systems, told Decrypt. Backing Carney's warning, Dadybayo said reliance on a few U.S. providers "creates genuine systemic risk, similar to what we saw in finance in 2008."

Distributing models across independent nodes removes the single "kill switch," but the risk persists if the compute behind them stays concentrated among a few suppliers, Dadybayo explained.

Such a trend is being driven by “rising compute and data costs,” Peter Anthony, founder and CEO of Perceptron Network, told Decrypt. Anthropic’s situation “didn't create that problem, it just made it impossible to look away from,” he added.

While Anthony agreed with Carney's point that the order marked a "strategic vulnerability," he questioned whether decentralization solves it or "just pushes the chokepoint back one layer to GPU suppliers."

If decentralized AI still runs on chips owned by a few cloud giants, "you've rebranded the risk, not removed it," he added.