This was not a good look for a company that prides itself on working closely with the academic community.
Anthropic is withdrawing a policy that hindered researchers from using its new Cloud Fable 5 LLM to build competitive AI models, the company reported. wired. “We are changing Fable 5’s security measures for Frontier LLM development to make them visible,” the company said in a statement. “We made the wrong compromise and we apologize for not getting the balance right.”
When Anthropic released the Cloud Fable 5, a new model based on its powerful Mythos system, researchers noted something strange. They found that when asked to perform certain tasks the Fable 5 would silently redirect requests to a smaller model. Furthermore, that restriction was not disclosed in the model’s documentation.
The new model was met with either dismissive or dismissive responses to tasks such as training competitive LLMs, debugging AI code, and optimizing neural architectures. Researchers were troubled not only by that decline but also by Anthropic’s lack of transparency about it. Of course, they were also worried that they had burned tokens and money on a model that didn’t live up to their expectations.
Anthropic has portrayed itself as a more ethical and researcher-friendly alternative to OpenAI, so its actions with Fable 5 provoked an intense reaction. Research fellow and Substack author Dean W. “The performance degradation on ML research without informing the user is shocking and horrifying,” Ball said on X.
Anthropic is not reversing its security policy on the Fable 5, but rather making the restrictions visible to users. “If the company suspects a user is trying to use the cloud to create a highly capable AI it will alert them, either denying the request, or redirect the user to a less capable model.” wired wrote.
<a href