Anthropic has apologized for secretly implementing throttling measures on its AI model, Claude Fable 5, through invisible guardrails, which have impeded users including researchers and competitors. The company announced it will now be more transparent about when these restrictions are activated, even if this results in Fable rejecting more queries.
We’re rolling out changes to make Fable 5’s safeguards for frontier LLM development visible.
Starting this week, flagged requests will visibly fall back to Opus 4.8—the same as our safeguards for cyber and bio. You will see this every time it happens. On the API, any flagged…
— ClaudeDevs (@ClaudeDevs) June 11, 2026
Fable is the inaugural model in Anthropic’s Mythos class of AI systems, which the company has cautioned could pose significant risks if released widely. In response to these risks, Anthropic launched Fable with built-in safeguards tailored to prevent it from answering certain “high-risk” queries, particularly relating to model distillation.
In Fable’s system card, Anthropic disclosed that it would modify and degrade the model’s answers if users attempted distillation without notification. Moving forward, queries identified as distillation attempts will automatically revert to Claude Opus 4.8, the preceding flagship model. Anthropic committed to informing users whenever their queries revert to Opus 4.8, stating, “You will see this every time it happens.”
This altered approach mirrors how Fable manages other high-risk queries, routing them through Opus 4.8 unless blocked by broader safety rules concerning topics like drugs or weapons. However, some restrictions have drawn criticism for being excessively broad, rendering Fable nearly unusable for basic queries in areas such as biology, as noted by Anthropic.
Anthropic acknowledged its initial decision for invisible safeguards was misguided, stating, “Invisible safeguards can be targeted more narrowly, allowing us to ship quickly with very few false positives … and that was the wrong tradeoff.”
The adjustments follow a significant backlash from the AI research community in reaction to Anthropic’s strategy of dynamically limiting users suspected of attempting to distill Fable for competitive purposes. In its system card, Anthropic justified the need to target such requests, explaining that using its models to create competing systems violates the company’s Terms of Service. The company has also accused certain competitors, including Chinese firms like DeepSeek, of unfairly diluting its models on an “industrial” scale.
