OpenAI, Anthropic, Google API Flaw Let Weaker AI Models Decode Stronger Models' Reasoning

A newly disclosed flaw in the way OpenAI, Anthropic, and Google carried hidden AI reasoning between API calls let researchers recover internal reasoning and secrets from session logs, including API keys and passwords. The weakness affected encrypted reasoning objects used by the providers' reasoning APIs, where a block created in one session could be replayed into another and, during testing,

The report describes a cross-provider API flaw in OpenAI, Anthropic, and Google reasoning systems that let researchers replay encrypted reasoning blocks into weaker models and recover hidden chain-of-thought content, including secrets such as API keys and passwords[1]. It also says the same weakness could expose hidden prompt injections and private data from shared logs, and that vendors have since mitigated the demonstrated attack path[1][2]. RealGround implication: organizations using reasoning APIs or publishing agent traces should treat opaque reasoning fields as sensitive data, remove them from logs and repositories, and review whether their AI workflows expose cross-session or cross-model replay risk.