Anthropic apologizes for secretly rerouting and degrading Fable 5 requests
Anthropic just apologized for quietly adding hidden safeguards to its new Fable 5 AI model.
It turns out some requests about cybersecurity, biology, or chemistry were secretly sent to a less advanced model (Claude Opus 4.8) without letting users know, and requests about AI development were quietly degraded.
The company admitted it made a "wrong trade-off," and critics were not happy about the lack of transparency.
Anthropic defends safeguards, promises clearer notices
Anthropic explained these measures were meant to prevent national security risks, like misuse in cyberattacks or dangerous bio projects.
After the backlash, it promised clearer notifications and explanations whenever requests are redirected or declined, so users will not be left guessing.
Most coding and machine learning tasks are not affected by this change.
Despite all the drama, Anthropic says Fable 5 is still its most powerful model yet: it can process a 50 million-line Ruby codebase a day and pull data from complex scientific charts with ease.