Anthropic launched Claude Fable 5, a new AI model described as a version of its Mythos system, but the rollout was immediately shadowed by two separate controversies.
According to The Decoder, Anthropic had been secretly throttling AI researchers who work for rival companies — quietly limiting what they could do with the tool without disclosing it. After the practice became public, the company acknowledged it had made a "wrong tradeoff" and reversed the policy. The Decoder notes that at least one other point of contention between Anthropic and the research community remains unresolved.
The timing of the launch added another wrinkle. TechCrunch reports that Fable 5 arrived just days after Anthropic had publicly warned that AI is becoming too dangerous — a striking contrast between the company's cautionary public messaging and its continued pace of releases.
Safety guardrails built into Fable 5 were challenged almost immediately after launch. An AI researcher claimed to have already bypassed them, according to TradingView. If verified, the claim would undercut one of the model's central promises.
Not all the news was turbulent. TechCrunch highlighted a more playful capability: Fable 5 can reportedly generate video games with a single button click, producing results described as "weirdly fun."
The episode matters because Anthropic has built its brand around being the safety-first AI lab — but secret throttling of researchers, a rapid guardrail bypass claim, and a launch timed days after its own public safety warnings put that identity under real scrutiny.