Anthropic has announced a revision to its policy regarding the Claude Fable 5 LLM, following criticism from researchers. The company stated that it will now make the safeguards affecting model performance clear to users, addressing concerns over transparency.
Researchers had reported that when utilizing Fable 5 for certain tasks, the model would redirect requests to a less advanced version without prior notice, leading to frustrations over wasted resources. Tasks impacted included training competing AI models and debugging code, which were met with degraded responses.
In light of its commitment to ethical practices and collaboration with the academic community, Anthropic acknowledged that their previous approach did not strike the right balance. While the restrictions will remain, users will now receive notifications when their requests are affected.