Anthropic's Claude API has blocked attempts to transcribe a public domain PDF of navigation rules due to its content filtering policy. The API error, a 400 Output blocked by content filtering policy, occurs when Claude is asked to reproduce existing text verbatim, regardless of licensing. This anti-regurgitation guard prevents Claude from outputting long spans that closely match source material, even for public domain documents like the USCG Navigation Rules and Regulations Handbook. The developer found that the model's intent or the text's public domain status did not bypass the filter, leading to a need for an alternative architecture that avoids routing source text through model output. AI
IMPACT Highlights limitations in current LLM APIs for tasks involving verbatim content reproduction, necessitating alternative architectural approaches for developers.
RANK_REASON The item describes a limitation and workaround for using an existing AI model's API, rather than a new release or significant industry event.
Read on dev.to — Anthropic tag →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
