OpenAI adds ChatGPT session controls, but governance gaps persist

OpenAI has added Active sessions to ChatGPT, giving users and administrators a new way to review and end logged-in sessions. The feature is available across all ChatGPT accounts and workspace types, including personal and managed workspaces, according to the company.

The update addresses a basic visibility gap that has long frustrated enterprise security teams. Ensar Seker, CISO at SOCRadar, said organizations previously had limited insight into where users were signed in and often had to rely on password resets or broad account actions to force re-authentication. “Granular session control is a more efficient and less disruptive approach,” he said. “From a governance perspective, session transparency improves accountability and supports investigations.”

Through Active sessions, users can see known browser and app sessions across ChatGPT, Codex, and API Platform. The interface shows device and browser details, approximate location, sign-in date and time, whether a device is trusted, and whether the session is current. Users can reach the feature under Settings > Security > Active sessions, where they can log out of specific sessions, remove devices from trusted services, or log out of all sessions.

OpenAI said logging out of all sessions can take up to 30 minutes to complete. The company also cautioned that session details may be “approximate or incomplete,” and the feature has limits. It does not show or manage connected apps or third-party app sessions, sign-ins through third-party services, Codex CLI sessions, or recently signed-out sessions.

The feature also does not apply to accounts linked to enterprise single sign-on, including SAML and OpenID Connect, according to OpenAI. That means some enterprise customers will still need to rely on their own identity and access controls for those accounts.

Security experts said the new controls are useful but basic. David Shipley of Beauceron Security said the ability to end active sessions is something that already exists on many platforms. “They should’ve had it sooner, but better late than never,” he said. He also said OpenAI could do more to stop ChatGPT from being used by threat actors to host malware, which he described as a current enterprise threat.

The larger problem, experts said, is not only session visibility but AI model change. OpenAI recently updated GPT-5.5 Instant in both the ChatGPT app and API to improve response style and quality, the company said. OpenAI said the update makes responses “more natural in everyday conversations” and “better paced in practical help tasks,” with fewer overly long or bullet-heavy replies.

That kind of ongoing model iteration creates a moving target for risk and compliance teams. Shipley said it is not sustainable to build a testing plan around a nondeterministic system. Seker added that many organizations evaluate a model before production, but when behavior changes within the same version family, earlier assumptions may no longer match real performance.

“The biggest governance challenge in AI is not model adoption, it’s model change,” Seker said. He said regulated industries are especially exposed because auditability, repeatability, and change management matter more there.

Valence Howden, advisory fellow at Info-Tech Research Group, said enterprises often cannot assess the implications of model updates against their policies, and sometimes are not even aware the updates happened. He said the lack of an opt-out before changes are incorporated means enterprises are effectively “red-teaming the updates with their clients.”

Seker said security teams are now being asked to manage fast-moving models, new features, and changing behavior while still maintaining compliance, risk management, and continuity. He said AI governance is difficult because organizations are no longer evaluating a fixed product, but a “continuously evolving service” where capabilities and integrations can change faster than traditional review cycles.

Both Seker and Howden said enterprises should treat AI models as living systems, not static software releases. That means continuous validation, ongoing monitoring, periodic reassessment, and explicit vendor change-management expectations around model updates and behavioral shifts. OpenAI’s new session controls improve visibility, but the governance problem remains tied to how often the model itself changes.