{"version":"2.1.282","anchor":"redacted-thinking-errors-are-now-classed-as-thinking-signa","canonical_anchor":"redacted-thinking-errors-are-now-classed-as-thinking-signa","heading":"Rejected redacted thinking data is now handled like a bad thinking signature","tier":"notice","area":"Thinking","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.282\/e\/redacted-thinking-errors-are-now-classed-as-thinking-signa","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.282","markdown":"### Rejected redacted thinking data is now handled like a bad thinking signature\n\nClaude Code now treats the API's \"invalid data in redacted_thinking block\" error the same way as a thinking signature error\n\n**Unclear.** What Claude Code does after classing the error, such as removing the thinking and retrying, is not confirmed.\n\n**What**\n\nThinking is the reasoning the model does before it answers. Some of it can come back in hidden form, called a redacted_thinking block. When the API rejects a request with an error whose message contains \"invalid data in redacted_thinking block\", Claude Code now classes it as a `thinking_signature` error. That is the same category it already used for errors about bad or missing thinking signatures, and it goes through the same handling. This applies to everyone.\n\n**Why**\n\nA session that hits corrupted redacted thinking data should now be handled the same way as one that hits a signature error, instead of failing outright.\n\n- Area: Thinking\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 1\/5"}