A US directive cutting off foreign access to Anthropic's two most powerful models has given Cohere the kind of opening enterprise AI companies rarely get: a live example of why sovereign AI is no longer just a policy slogan.
On June 12, Anthropic received a US government export control directive ordering it to block foreign nationals from using Claude Fable 5 and Mythos 5, including foreign national employees inside the company. WIRED reported that the letter arrived at 5:21pm ET on Friday and did not spell out the specific national security concern. Anthropic's answer was blunt. It took both models offline for customers rather than try to verify nationality across every user and deployment in real time.
The shutdown immediately changed the sales conversation for companies that have spent the past two years arguing that AI buyers need alternatives to US frontier labs. Bloomberg reported on June 15 that Cohere described the inbound interest from enterprise customers as huge. That is a useful word for the Toronto company, but it is also a loaded one. Interest is not the same as migration, and a Friday-night procurement panic is not the same as signed annual recurring revenue.
Still, the event gave Cohere a cleaner argument than any white paper could. Enterprise buyers do not change model providers quickly. They move when performance fails, costs get out of hand, or risk becomes too visible for the board to ignore. A model disappearing from production because of a government directive sits squarely in that last category. It turns sovereignty from an abstract compliance preference into an operational question: can the tool your team depends on be removed by a regulator in another country?
That is the case Cohere and Aleph Alpha were already making before this week. In April, Cohere agreed to acquire Germany's Aleph Alpha in a deal backed by a $600 million investment from Schwarz Group, according to the companies and reporting cited by The New York Times. The combination was pitched around sovereign AI for governments and large enterprises, with Aleph Alpha bringing German public-sector relationships and Cohere bringing enterprise language models that can run in private cloud or customer-controlled environments.
Aidan Gomez, Cohere's co-founder and CEO, framed the deal at the time as a way to deliver sovereign AI to nations around the world. Two months ago, that sounded like strategic positioning in a crowded market. After Anthropic's Fable 5 and Mythos 5 shutdown, it reads less like branding and more like a sales deck with a real incident attached.
The sale is still the hard part
Cohere is not alone in seeing the opening. Mistral has been making a similar pitch from Paris with open-weight models and enterprise deployment options. OpenAI will argue from the other direction, that its models were not hit by the directive and that market dominance has its own kind of reliability. For buyers, this is no longer a simple contest between model benchmarks. The procurement file now has a new column for political and regulatory exposure.
The Wall Street Journal reported that the US action followed concerns raised by Amazon researchers and CEO Andy Jassy about whether Fable 5 could be prompted to produce cyberattack-related information. Anthropic has disputed the severity of the issue. AP reported that the company called the order a misunderstanding and said it hoped to restore access as soon as possible. The Commerce Department did not immediately comment to AP.
That caveat matters because Cohere's window may be short. If Anthropic restores access within days, some buyers will treat this as a frightening outage rather than a reason to rebuild their AI stack. Large companies do not swap foundation model providers over a weekend. They run evaluations, security reviews, legal checks, and internal politics. The more deeply a model is woven into workflows, the harder it is to remove cleanly.
The better play for Cohere is to sell resilience rather than outright replacement. A financial institution in Frankfurt or a ministry in Ottawa may not want to abandon Anthropic, OpenAI, or Google. It may want a second model path, one that can sit inside its own infrastructure and keep critical work moving if a foreign directive interrupts access again. That is a smaller first contract than a full displacement deal, but it is more believable and easier to justify after this week.
The strongest part of Cohere's position is that it does not need to persuade buyers that geopolitical AI risk could someday exist. The example is already on the table. Anthropic received a Friday letter, two major models went dark, and customers around the world had to explain what happened. Even if Fable 5 and Mythos 5 return quickly, that memory will sit inside procurement teams for a long time.
For Gomez and his team, the question is execution. Inbound interest can flood a sales team and still leave little behind if it is not turned into contracts, pilots, and deployment plans before the urgency fades. Cohere has the sovereign AI story it wanted. Now it has to prove it can make that story useful before every rival starts telling the same one.
Also read: The Pentagon just ran the fastest enterprise AI rollout in history and it is only half done; NewCore launches with $66 million to give AI agents proper enterprise identities; Nvidia sells $20 billion in bonds as AI demand outpaces its own cash generation
