both openai and anthropic actually do this and let you as a harness store only the encrypted contents of thinking traces to be passed again on further turns but

both openai and anthropic actually do this and let you as a harness store only the encrypted contents of thinking traces to be passed again on further turns but they only have the key to unencrypt it so its jibberish for final users https://developers.openai.com/cookbook/examples/responses_api/reasoning_items#encrypted-reasoning-items https://developers.openai.com/cookbook/examples/responses_ap...