Les États-Unis ont levé les restrictions sur le modèle Fable and Mythos 5

The United States has lifted export restrictions on two advanced AI models developed by Anthropic – Fable 5 and Mythos 5 – marking a significant shift in the regulatory landscape surrounding artificial intelligence. This decision, made three weeks after initial limitations were imposed, comes amid growing competition between American and Chinese tech firms, particularly in the field of large language models. The move signals a recalibration of U.S. policy toward AI innovation, balancing national security concerns with the need to maintain technological leadership in the global race for AI supremacy.

The lifting of export controls was announced by the U.S. Department of Commerce following internal assessments that had initially flagged potential vulnerabilities in the Fable 5 model. These vulnerabilities, identified by researchers at Amazon, raised concerns about how the model could potentially bypass certain security measures. As a result, the U.S. government temporarily restricted access to Fable 5 for international users, citing risks to national security. However, after further evaluation, officials concluded that the initial assessment had overestimated the threat, leading to the reversal of these restrictions.

Anthropic, which develops both Fable 5 and Mythos 5, expressed mixed reactions to the decision. While the company welcomed the return of its models to global markets, it also noted that some security features had been modified to address the concerns raised during the review process. These changes have implications for specific tasks related to programming and error resolution, with some requests now being automatically redirected to the Opus 4.8 model as a safer alternative. For enterprise customers using paid plans, Fable 5 will remain available until July 7th, but usage costs will increase significantly compared to Opus 4.8, reflecting the heightened security measures and resource demands associated with the updated model.

The timing of this decision coincides with intensifying competition between U.S.-based AI developers and their counterparts in China. Reports suggest that Chinese models have already achieved comparable performance in cybersecurity testing against Anthropic’s Mythos 5, while offering more cost-effective solutions. This competitive dynamic has prompted broader discussions within U.S. technology circles about the long-term strategic implications of such regulatory decisions. A recent episode of the *Hard Fork* podcast, produced by *The New York Times*, highlighted how the removal of export restrictions for Anthropic’s models reflects a larger trend in U.S. AI strategy—balancing national security interests with the imperative to stay ahead in the global AI arms race.

In addition to Anthropic, other major players in the AI space are also facing similar scrutiny. The podcast suggested that similar restrictions might soon apply to OpenAI’s upcoming GPT 5.6 model, indicating that the U.S. government is likely to adopt a more nuanced approach to regulating AI exports in the future. This evolving regulatory environment underscores the complex interplay between innovation, security, and geopolitical competition in the rapidly expanding field of artificial intelligence.

For Anthropic, the reinstatement of global access to Fable 5 and Mythos 5 carries substantial business significance. The company is currently preparing for its anticipated initial public offering (IPO), and increased access to international markets is seen as crucial for boosting revenue and strengthening its position in the global AI industry. As the AI sector continues to evolve, the balance between regulation and innovation will remain a central theme in shaping the future of artificial intelligence development worldwide.