Data Exfil: OpenAI polices private chats exfiltrated through Codex Memories (github.com)

OpenAI's Codex has come under scrutiny for allegedly exfiltrating private chat content from users' local providers without consent. A recent investigation revealed that the memory function could send serialized transcripts from one provider to OpenAI’s systems, even if the source was a non-OpenAI provider. This data transmission occurred without user knowledge, triggering concerns about privacy and data security. The journalist behind the findings highlighted that OpenAI's actions led to account moderation warnings based on content that should have remained private and under user control. This scenario raises significant questions for the AI/ML community regarding ethical data handling practices. OpenAI's apparent ability to cross-reference and process data across different providers could set a concerning precedent for user consent and privacy in AI interactions. While the current setup allows for potentially useful memory generation, it also risks breaching user confidentiality if not transparently managed. The implications call for clearer user agreements, stronger data privacy measures, and rigorous audits of OpenAI's practices to ensure responsible AI deployment.