The biggest winners and losers from US restrictions on Anthropic's AI

The recent decision by the White House to impose restrictions on Anthropic's AI models has sent ripples through the industry, creating distinct winners and losers. On Friday, US officials limited access to Anthropic's cybersecurity-focused systems, Mythos 5 and Fable 5, after concerns arose over the potential for safeguards against misuse to be circumvented. This move has not only restricted foreign nationals from accessing these systems but has also forced Anthropic to shut down access entirely for all users. This development has posed a significant challenge for Anthropic, but it simultaneously opens doors for other AI companies that emphasize more open models. For instance, French startup Mistral, which has advocated for European independence from American AI suppliers, finds itself in a favorable position. Unlike Anthropic, Mistral offers open-weight models that clients can deploy on their infrastructure and customize using their own data. The recent restrictions serve as a real-world example of the risks associated with relying on closed models, prompting Mistral's CEO, Arthur Mensch, to reinforce the importance of model sovereignty in a recent social media post. The timing of Mistral's rise aligns perfectly with France's announcement that its intelligence agency will switch from using Palantir's AI tools to those provided by a domestic company. Prime Minister Sébastien Lecornu highlighted the need to avoid "strategic dependencies" on foreign technology, further validating Mistral's approach. Similarly, DeepSeek stands to benefit from the situation. The company's open-weight models allow users to download, modify, and deploy their systems independently, making them more attractive as governments and businesses seek greater control over their AI solutions. This scenario also offers China a chance to argue against reliance on US AI providers, framing it as a geopolitical risk, especially as Chinese AI models are perceived to lag behind their American counterparts by about six to twelve months. Conversely, Anthropic faces significant setbacks due to these restrictions. Not only do the export controls limit the ability of foreign nationals, including its own employees, to access critical systems, but they also spotlight the vulnerabilities inherent in closed AI models. With Anthropic controlling access to its technologies, it leaves itself susceptible to governmental influence over who can utilize these systems. This episode further complicates Anthropic's ongoing tensions with the White House, stemming from its assertion that its technology should not facilitate domestic mass surveillance or fully autonomous weapon systems. The US government has classified Anthropic as a supply chain risk, leading the company to challenge this designation in court. While Anthropic's US competitors may experience short-term gains from these restrictions, the long-term implications could be troubling for them as well. Companies like OpenAI, Google, and xAI predominantly deliver their advanced models via their own controlled platforms, raising questions about their future accessibility. In contrast, Meta, which has some open-weight models, is increasingly leaning towards tighter control with models like Muse Spark. The Anthropic situation serves as a critical reminder for governments and businesses alike: access to AI technologies can hinge on the decisions made by providers and the regulatory frameworks governing them. This dynamic could amplify the appeal of sovereign and open-weight alternatives across Europe and beyond.