Anthropic's AI model Fable 5 has run into a government roadblock. According to Wired's Hugo Lowell, Trump administration officials say that if Anthropic wants to rerelease Fable 5, the company will first need to ensure the model's guardrails cannot be circumvented.
Guardrails are the built-in safety limits that are supposed to stop an AI model from producing harmful or prohibited content. The administration's position, as relayed to Wired, is essentially a precondition: prove the protections will hold, or the model stays off the market.
There is a catch. Experts cited by Wired say that making guardrails truly impossible to bypass may not be achievable. In practice, determined users and researchers have repeatedly found ways around the safety controls on large AI models, which makes an absolute guarantee a difficult — possibly unrealistic — bar to clear.
The dispute also raises a broader question about how the government is treating AI models. A report from cio.com argues that the Anthropic–Fable situation suggests the word "export" no longer means what it used to, hinting that regulators are starting to treat the release of an AI model less like shipping software and more like a controlled good subject to government conditions.
Why it matters: if Washington can demand unbreakable safeguards as a condition of releasing an AI model, it sets a precedent that could reshape how — and whether — powerful models reach the public.