Anthropic has released Claude Fable 5, its first model in what the company calls the "Mythos class" — a tier it considers its most capable AI to date. According to TechCrunch, the release came just days after Anthropic publicly warned that AI is becoming too dangerous. The model has drawn attention for at least one flashy capability: TechCrunch reports it can generate playable video games with minimal user input.
But the launch arrived with two significant controversies in tow.
Almost immediately after release, an AI researcher claimed to have already bypassed Fable 5's safety guardrails, according to TradingView. The claim struck at the core of Anthropic's reputation as a safety-focused lab.
A separate and arguably more damaging story involves how Anthropic treated researchers from competing companies. According to Engadget, the company had been running a hidden policy that researchers say "sabotaged" their work by throttling their access. Anthropic has since reversed course. The Decoder reports the company admitted it had made the "wrong tradeoff" by invisibly throttling rival AI researchers — a rare public concession from a major AI lab.
A third dispute centers on data. A report flagged widely on Hacker News, and confirmed by The Verge, shows Anthropic now requires a 30-day data retention period for Fable and Mythos-class models. That requirement has rattled at least one major enterprise customer: Analytics India Magazine reports Microsoft has restricted internal use of Claude Fable 5 over data retention concerns, while Techzine Global notes Microsoft is exploring how to adapt its own retention policies to work around the constraint.
The cluster of controversies matters because it exposes a widening gap between the AI industry's public safety messaging and the less visible decisions — about who gets access, on what terms, and how long data is held — that will ultimately determine who can scrutinize these systems and how.