OpenAI's Codex now watches your screen to remember what you're working on

The security angle here is underplayed: Chronicle doesn't just retain session context, it watches the screen continuously, meaning credentials, API keys, and proprietary logic that a developer never explicitly shares with the assistant could be ingested passively and stored. That's a materially different risk profile than opt-in memory.

This is the next increment in a rapid capability push that started with the April 16 Codex expansion covered by TechCrunch and The Verge, where OpenAI added computer control, image generation, and memory to Codex in a direct response to Anthropic's Claude Code. Chronicle looks like the memory layer from that update being operationalized at the screen level. The competitive framing from Stratechery's coverage of OpenAI's internal memo targeting Anthropic in enterprise markets is relevant here too: persistent, ambient context is exactly the kind of stickiness that matters in B2B developer tooling, where switching costs compound over time. OpenAI's own cybersecurity program, announced the same week, makes the data-retention risk harder to dismiss as theoretical.

Watch whether enterprise security teams or major cloud providers issue formal guidance on Chronicle within the next 60 days. If they do, OpenAI will face pressure to offer an auditable data-retention policy or an air-gapped deployment option, which would test how serious the B2B push actually is.