Skip to content
Submit incident
AII-1000-47341
ReportedReported by fewer than three independent publishers, with no official finding.

ChatGPT Work gains autonomous account login capability raising security concerns

OpenAI launched a feature in ChatGPT Work on 25 August 2026 that allows the AI agent to autonomously log into and remain authenticated in user web accounts using stored session cookies. Security researchers raised concerns that persistent authentication sessions in a remote cloud environment could pose risks if the system is compromised or misused.

25 August 2026 Β· OpenAI Β· Security and cyberattacks
25 August 2026
AII-1000-473412 sources ↓
Quick facts
Organization
OpenAI (subject)
Category
Security and cyberattacks
Country
United States
Source
Notebookcheck
Date
25 August 2026
Drafted by
Raidu Incident Detection

What happened

On 25 August 2026, OpenAI extended ChatGPT Work's cloud browser to handle login-protected websites. When an agent encounters a login page, users enter their username and password through a secure form, and the browser then maintains an authenticated session using cookies for future tasks without requiring re-entry of credentials.

According to OpenAI's documentation, the cloud browser runs on a remote machine independently and continues operating even after the user closes their device. The session cookies persist in this remote browser, allowing the agent to access authenticated accounts while the user's device is offline.

Reported impact

Security researchers raised concerns that the persistent session cookies stored in the cloud browser could potentially allow unauthorized access to user accounts. A cookie containing authentication information is sufficient to access an account without the original password or two-factor code if the remote browser is compromised.

Index analysis

The feature relies on users to restrict which sites the cloud browser can access through settings with three options: 'Always ask', 'Auto approve', or 'Always allow'. OpenAI explicitly notes that 'Always allow' is 'not recommended'.

OpenAI stated that credentials entered through the secure form go directly to the remote browser and are not stored, seen by the model, or used for training. However, the authentication session and associated cookies remain active in the cloud browser environment.

Classification

Organization
OpenAI (subject)
Category
Security and cyberattacks
Industry
Digital security
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Unknown
Failure modes
Unsafe autonomous behaviour
Consequences
None recorded
AI systems
ChatGPT Work

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

Notebookcheck
ChatGPT can now sign in to your accounts and stay signed in
26 August 2026
ZDNet
ChatGPT can log into your web accounts without you now - but should you let it?
27 August 2026