TechCrunch Publishes Practical Guide to Detecting Compromised ChatGPT, Claude, and Perplexity Accounts

As AI chat assistants increasingly hold sensitive context, including saved conversations, connected tools, and memory of personal and work details, account takeover on these platforms has become a real exposure, and TechCrunch published a walkthrough of how to check whether accounts on ChatGPT, Claude, or Perplexity have been accessed by someone else. The practical value for builders and IT-security practitioners is in the platform-specific mechanics, since each service exposes session visibility differently. ChatGPT surfaces an Active Sessions panel under Settings, Security and Login, letting a user spot unfamiliar devices and revoke either a single session or all of them, and it supports traditional password resets alongside multi-factor authentication. Claude, by contrast, has no password at all; Anthropic authenticates purely via emailed login links, so its equivalent Active Sessions list under Settings, Account is the only lever available to terminate suspicious access, and there is no password-reset step to fall back on. Perplexity is the least transparent of the three: it does not expose a session list, so the only remediation path is a blanket sign-out of all sessions, followed by re-authentication via a one-time email code. For engineering and security teams standardizing incident-response playbooks or writing internal security-awareness guidance, this is a useful reference because AI platforms are increasingly treated as high-value targets. Attackers who gain access can exfiltrate proprietary code shared in chat history, pivot through connected integrations, or impersonate the account holder in agentic workflows. The piece is a reminder that account-security hygiene, such as unique passwords in a manager and MFA where available, needs to explicitly extend to AI tooling, not just traditional SaaS and email accounts, and that incident-response steps differ meaningfully across providers because their authentication models are not standardized.

Source

View on ShipDigest