OpenAI brings Codex into Chrome

Browser automation is the point here — not just a new extension badge in Chrome. OpenAI added a Codex extension on May 7 that lets its coding agent work inside your actual browser session, including sites where you’re already signed in. That matters because a lot of modern software work breaks at the browser boundary. The code may compile fine, but the real test lives in the app, the admin panel, the PR page, the dashboard, or the bug report you can only see after login. (developers.openai.com) ### What actually launched? OpenAI launched “Codex for Chrome,” a new extension documented in its developer docs and listed in the Codex changelog on May 7, 2026. The extension lets Codex use Chrome for tasks that need your signed-in browser state, while the rest of the Codex app keeps handling code, terminals, files, and other agent workflows. (developers.openai.com) ### Why does Chrome matter so much? Because OpenAI already had an in-app browser — but that browser is deliberately limited. It does not support your normal browser profile, existing tabs, cookies, extensions, or authenticated flows. That makes it good for localhost previews and public pages, but bad for the messy, (developers.openai.com)l or an internal web app. The Chrome extension fills exactly that gap. (developers.openai.com) ### What can Codex do in Chrome? The short version is: use the web where you already work. OpenAI says Codex can read from or act on signed-in sites like LinkedIn, Salesforce, Gmail, or internal tools, and it runs those browser tasks in Chrome tab groups so each thread stays contained. The changelog also says it works across(developers.openai.com)tle but important design choice — the agent is meant to operate alongside you, not hijack your active session. (developers.openai.com) ### How much control does the user keep? Quite a lot, at least by design. Codex asks before interacting with a new website by default, and users can allow a site for one chat, always allow a host, or block it. There are also allowlists and blocklists in settings. That matters because once an agent can touch your rea(developers.openai.com)bad clicks, bad context, or sensitive data leaking into a task. (developers.openai.com) ### What’s the catch? The catch is context contamination. OpenAI explicitly warns that page content should be treated as untrusted, and that browser history can contain sensitive telemetry, internal URLs, search terms, and activity from signed-in Chrome sessions. If a user allows history access, relevant entries can(developers.openai.com)useful also makes permissions and review much more important. (developers.openai.com) ### How does this fit into the bigger Codex push? It looks less like a one-off extension and more like the next step in a broader product shift. In April, OpenAI said Codex had grown into a wider workspace with computer use, an in-app browser, PR review support, SSH access, multiple terminals, and more than 90 plugi(developers.openai.com)to the actual tools and web apps developers use all day. (openai.com) ### Why should developers care? Because browser validation is where a lot of engineering time disappears. OpenAI has already described wiring browser tooling into Codex’s workflow to reproduce bugs, inspect UI state, and validate fixes, and it says internal teams used Codex to drive a codebase that reached about 1 million lines a(openai.com)stic: write code, open the app, inspect the result, review the PR, and keep going without switching to a separate human-only step every time. (openai.com) ### Bottom line? This is really a story about where the coding agent lives. Codex started as something that wrote and reviewed code in controlled environments. Now OpenAI is pushing it into the browser — the place where software actually gets checked, approved, and used. If that works reliably, the browser stops being the handoff point and starts becoming part of the agent’s workspace. (developers.openai.com)