ChatGPT's New Agent Mode Raises Questions About Automated Account Access
OpenAI has begun rolling out agentic capabilities for ChatGPT Work that allow the AI to sign into users' online accounts automatically, without requiring human interaction during the login process. This feature is designed to streamline workflows by letting ChatGPT handle tasks across third-party services on a user's behalf.
The development represents a notable step in the broader trend of AI agents — systems that can take actions, not just generate text. By storing and using credentials to access external services, ChatGPT can theoretically browse, fill forms, read data, or perform multi-step tasks within other platforms.
Security and privacy researchers have flagged several considerations. Granting an AI system persistent access to personal or corporate accounts introduces new attack surfaces. If the underlying system is compromised, or if access tokens are mishandled, the consequences could extend well beyond a single session. Questions also arise around data retention: what OpenAI logs, how long it retains session data, and whether third-party services' terms of use permit automated access by AI agents.
For enterprise deployments, IT and security teams may need to evaluate whether such capabilities comply with their organization's authentication policies and risk frameworks. Users, for their part, face a tradeoff between the convenience of hands-off automation and the potential exposure that comes with handing over login credentials to an external AI service.
OpenAI has not publicly detailed the full scope of safeguards built into this feature, though the company is expected to expand documentation as the rollout progresses.