Anthropic says it is changing how its AI handles certain prompts tied to national security. According to reporting, the company previously downgraded or rejected some user requests without clearly informing users of the reason, which drew backlash. In response, Anthropic now says its system will tell users when their request is rejected or downgraded specifically for national security concerns. The update aims to make the moderation decision more transparent, rather than leaving users to infer why a request received a reduced answer quality or was not fulfilled.
The disclosures described in the coverage focus on user-facing communication around refusals or downgrades, not on a specific list of prohibited topics. Both outlets characterize the shift as a reaction to criticism of the earlier behavior, emphasizing that the new approach explicitly labels national security as the basis for the decision when applicable.
