Anthropic this week launched Claude Fable 5, the first publicly available model in its new Mythos family and what the company describes as its most powerful AI model yet. According to reporting by MSN, the model is said to outperform rivals including ChatGPT and Gemini across a variety of tasks — a bold claim that put it immediately in the spotlight.
But the launch quickly turned turbulent. According to The Verge, Anthropic stealthily embedded hidden guardrails in Fable 5 that throttled the model's capabilities — specifically in ways that undermined researchers and competitors attempting to use it to build rival AI systems. The practice, sometimes called distillation blocking, was not disclosed to users up front.
Facing backlash, Anthropic apologized and said it is reversing course. The company also pledged to be more transparent going forward about when and how such restrictions activate, according to The Verge.
The controversy didn't stop there. According to Moneycontrol, Microsoft warned its own employees not to use Claude Fable 5, citing an internal message that read "Legal is still evaluating Anthropic's" terms — suggesting Microsoft's legal and compliance teams had unresolved concerns about the model's usage policies at the time of launch.
The episode matters because it surfaces a tension that is only going to intensify: AI companies want their models widely adopted, but are increasingly tempted to quietly limit how that adoption works — especially when a competitor might benefit. Transparency about those limits, it turns out, isn't optional.