This week's Anthropic-inspired AI freakout, explained

In a significant shift following backlash from developers, Anthropic has announced changes to its AI model usage policy. The company will no longer obscure the performance of its Fable 5 model in response to developer inquiries. Instead, requests for assistance in frontier AI model development will now be directed to the less advanced Opus 4.8 model, with developers informed of this adjustment. This move aims to quell concerns that Anthropic was intentionally providing inferior responses while misleading users about the quality of its outputs. Despite the apology and policy change, Anthropic continues to impose restrictions on the use of its most advanced public model for specific AI development tasks. The company justifies these limitations by citing safety concerns, claiming they prevent foreign adversaries from leveraging Anthropic’s top-tier model to undermine the United States’ competitive edge in AI and semiconductor technology. However, this rationale has its critics, as it also serves to protect Anthropic’s business interests from potential intelligence extraction by competitors. This phenomenon, where rivals harness the outputs of a powerful model to enhance their own systems, poses a threat not just from foreign entities but also from open-source developers in the U.S. and Europe. The implications of these restrictions are vast, as they apply to anyone working on competing AI models. Anthropic's terms of service explicitly prohibit the use of its products to develop rival offerings, equating Western open model developers with foreign adversaries. This blurs the line between national security and business strategy, raising questions about the true motivations behind these limitations. Open models have emerged as formidable competitors. A study from MIT Sloan highlighted that open models achieve about 90% of the performance of closed models and have been closing the performance gap at an accelerating rate. Recent data from Artificial Analysis shows that while Anthropic's models excel in certain tasks, open-source alternatives like Xiaomi's MiMo v 2.5 Pro are quickly catching up, often at a fraction of the cost. For instance, Xiaomi's model charges just 43 cents per million input tokens, in stark contrast to Anthropic's $10 per million. This price disparity, coupled with comparable performance, poses a significant challenge for Anthropic as it continues to invest heavily in developing cutting-edge models. Experts like Nicholas Vincent, a computer science professor, suggest that the company's actions may reflect more of a business strategy than a purely safety-driven approach. As Anthropic navigates these complexities, transparency about the interplay between safety and commercial interests will be crucial as it faces growing competition in the AI landscape.