Summary
- Canadian Prime Minister Mark Carney highlighted the dangers of relying on a limited number of AI providers, citing U.S. interference with Anthropic.
- The U.S. government mandated Anthropic to disable two of its advanced AI models globally.
- While decentralized AI mitigates the "kill switch" risk, it still depends on a few chip manufacturers, experts noted.
Mark Carney, Canada's Prime Minister, expressed concerns on Sunday regarding U.S. restrictions imposed on Anthropic's cutting-edge AI models, emphasizing the dangers of depending on a small group of American companies after a directive led to the global shutdown of two advanced systems.
"The current situation with Mythos and Fable illustrates the risks associated with overreliance on specific models," Carney stated during his remarks in Ireland ahead of the G7 summit in France, according to the Associated Press.
While Carney acknowledged that no party had acted improperly, he cautioned against complacency, stressing the importance of learning from the incident and advocating for diversification in AI development.
“Relying on a single option is never advisable,” he pointed out.
His comments followed a government order issued on Friday which required Anthropic to restrict access to its Fable 5 and Mythos 5 models for foreign users, citing national security concerns.
In response, Anthropic disabled both models for all clients while contesting the rationale, arguing that the jailbreak mentioned is already replicable on publicly available models like OpenAI's GPT-5.5.
According to reports, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei regarding the directive, which was partly motivated by suspicions that a group linked to China had accessed Mythos, as noted by Semafor.
As Anthropic approaches a $1 trillion valuation with an annual revenue exceeding $47 billion, Decrypt has reached out to the company for further comments and will update this article if they respond.
Challenges and Implications
Carney's observations frame centralized AI as a potential single point of failure, where compliance with governmental directives can restrict access for users globally.
Centralized AI keeps its models and controls within one firm, while decentralized AI distributes these functions among multiple independent operators coordinated through a blockchain.
Following the restrictions on Anthropic's models, decentralized AI projects saw a market cap increase to $24.3 billion, reflecting a 6% rise for the day and a 12% increase over the week, according to CoinGecko.
Smaller computing and data networks led the gains, with projects like ChainOpera AI, io.net, Grass, and NOVA surging by over 30% in the past week, while major tokens such as NEAR Protocol and Bittensor increased by 15.9% and 27.9%, respectively.
Dan Dadybayo, strategy lead at Horizontal Systems, highlighted that the U.S. government's action regarding Anthropic underscores a risk unique to centralized AI. Supporting Carney's caution, he indicated that reliance on a few U.S. providers creates significant systemic risk, reminiscent of the financial crisis in 2008.
While decentralized models mitigate the risk of a single "kill switch," Dadybayo noted the potential for risk to persist if computing resources remain concentrated among a limited number of suppliers.
Peter Anthony, founder and CEO of Perceptron Network, attributed this trend to escalating costs associated with computing and data. He remarked that Anthropic's situation did not create this issue but made it impossible to ignore.
Anthony concurred with Carney's assertion that the order represents a "strategic vulnerability," but he questioned whether decentralization effectively addresses the issue or merely shifts the chokepoint to GPU suppliers. "If decentralized AI still depends on chips owned by a few cloud giants, it merely rebrands the risk without eliminating it," he cautioned.