Anthropic’s AI restrictions caused concern among founders, researchers and developers following the release of the Mythos-like model.
It was just this week that Anthropic launched Claude Fable 5, an AI model that the company described as “Mythos-class”, but with additional significant security restrictions in place to prevent misuse.
Already, however, the organisation is facing backlash from founders, researchers and developers who find the ‘secretive’ large language model (LLM) policies to be deliberately limiting competitors and users in the development of alternative AI models.
In its statement announcing the availability of Claude Fable 5, Anthropic explained that the model has barriers designed to block responses that stray into high-risk areas – notably cybersecurity, chemistry and biology. Such interactions, it stated, would instead be rerouted to Opus 4.8, a less powerful model.
This has reportedly drawn some criticism from researchers and developers in the AI space, as there is a concern that work derived this way is deliberately degraded in a manner that is invisible and secretive.
Anthropic has since responded to the backlash, indicating plans to address the issue by making Claude 5’s safety rails visible to all users. If the company suspects that a user is attempting to develop their own high-powered AI system, it will send a clear prompt informing them that the request is being refused based on policy, or that it is being rerouted to a less capable model.
It is currently against Anthropic’s terms of service to use Claude technology as a means of training competing AI models.
In a statement to Wired, a representative for Anthropic said, “We’re changing Fable 5’s safeguards for frontier LLM development to make them visible. We made the wrong trade-off and we apologise for not getting the balance right.”
Among the critics of Anthropic’s decision-making was Dean Ball, a senior fellow at the Foundation for American Innovation and a former adviser to the White House on AI. In a post published on X, Ball described deliberately degrading “ML research” performance without informing the user as a “shockingly hostile and terrible look”.
He further explained its potential to “silently damage all sorts of work” and “raise the eyebrows of antitrust enforcers worldwide.”
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
You must be logged in to post a comment Login