June 11, 2026
Now You See It, Now You Don’t
Anthropic apologizes for invisible Claude Fable guardrails
Anthropic got busted for secret AI limits, and the internet is absolutely not buying the apology
TLDR: Anthropic admitted its new chatbot was secretly weakening some answers and now says it will show users when that happens instead. Online, many people aren’t buying the apology, arguing the bigger issue is trust: an AI that quietly changes replies is hard to rely on and looks suspiciously self-serving.
Anthropic thought it could quietly slip a hidden brake pedal into its new AI model, Claude Fable, and users would just… not notice. Instead, the company is now in full apology mode after admitting Fable was secretly changing or weakening answers when it suspected people were using it to help build rival AI tools. After the backlash, Anthropic says it will stop doing this invisibly and will now clearly tell users when a request gets handed off to an older model instead.
But the real fireworks are in the community reaction, where people are split between "safety measure" and "sneaky moat defense." One of the strongest complaints was simple: if the AI won’t answer, it should fail cleanly instead of quietly messing with replies and leaving users guessing. Others were much harsher, accusing Anthropic of only being sorry because it got caught, not because it did anything wrong. That suspicion fueled the biggest drama: was this about protecting the public, or protecting the company’s competitive edge?
And yes, the jokes arrived right on schedule. One commenter snarked that the article itself sounded like it was "written by Claude and forwarded to Verge," while another boiled the whole scandal down to a brutally memeable line about Anthropic apologizing for "defending their moat." In other words: the company wanted invisible guardrails, but what it got was very visible distrust.
Key Points
- •Anthropic said it will stop silently degrading Claude Fable 5 outputs for suspected distillation-related prompts.
- •Claude Fable 5 is described as the first widely available model in Anthropic’s Mythos class, which the company had previously warned was too dangerous for public release.
- •Fable’s system card said suspected distillation attempts would trigger altered and degraded responses without informing users.
- •Anthropic now says those queries will instead be routed to Claude Opus 4.8 and users will be explicitly notified each time.
- •The article says this revised handling matches Fable’s existing treatment of other high-risk areas such as biology, chemistry, and cybersecurity.