OpenAI’s Codex is no longer just a coder’s helper — and users are pushing back

Lead

OpenAI’s Codex is being repositioned from a code-generation assistant into a cross‑application agent that can operate your desktop and carry work forward across days. That ambition has impressed many — and alarmed others. It has been reported that the latest update gives Codex the ability to “see the screen, click the mouse and type on the keyboard,” run parallel background agents, and remember user preferences so tasks can resume automatically over time.

What changed

The upgrade bundles previously scattered capabilities into a single workflow engine: native macOS desktop control, browser‑level UI annotations, SSH access to remote environments, multi‑terminal management, GitHub review handling, side‑bar previews for PDFs and spreadsheets, and a new built‑in image generator (gpt-image-1.5). It also expands plugin integrations — reportedly more than 90, including JIRA, GitLab and Microsoft Office — and OpenAI says Codex now has over 3 million weekly users with almost half using it for non‑coding tasks. In short: rather than replying to prompts, Codex can now act across apps and time to complete work for you.

Why many are worried

Should users worry? Privacy and security advocates say yes. Allowing an AI agent to interact with local apps, read messages in Slack, Gmail and Notion, and persist context across sessions raises questions about data leakage, consent and corporate compliance. Developers and IT teams also warn about potential automation errors, unwanted changes to internal systems, and vendor lock‑in when a single “super app” orchestrates many disparate tools. It has been reported that the desktop control feature is initially macOS‑only and that memory and advanced context features are rolling out first in the US, with the EU, UK and enterprise/education editions coming later — a staggered launch that reflects both product testing and regulatory caution.

Rollout and the regulatory backdrop

OpenAI’s move comes as regulators worldwide sharpen scrutiny of AI systems that handle personal and business data. GDPR, corporate security policies, and geopolitical tensions around technology supply chains mean enterprise customers will be watching how data is stored, shared and audited. Reportedly, OpenAI plans to broaden browser control and deeper integrations over time, but whether users and corporate buyers will accept an AI that effectively embeds itself in daily workflows remains an open question.