Fake OpenAI Tenant Invites Used to Steal Sensitive Company Data
Threat actors are exploiting the trusted reputation of OpenAI by creating fraudulent organizational tenants and inviting employees to join, mimicking legitimate company identities to appear credible. Because the invitations arrive through official OpenAI notification channels and include convincing details like attached credit cards, standard email security controls fail to flag them as suspicious. Employees who accept these invites may unknowingly submit sensitive assets — such as source code or customer data — directly into attacker-controlled ChatGPT environments. This attack highlights how the rapid adoption of AI tools has outpaced organizational policies governing their safe use, creating a dangerous gap between employee behavior and security oversight.
Tactical Insight
Immediate actions
- Audit and inventory all AI platform accounts (including OpenAI, Microsoft Copilot, etc.) to identify any unauthorized organizational memberships.
- Instruct employees to verify any AI platform organization invites through official internal IT channels before accepting.
- Issue a company-wide alert warning staff of this specific 'Poisoned Tenant' social engineering technique.
Long-term improvements
- Establish and enforce an Acceptable Use Policy (AUP) that explicitly governs what data employees may submit to AI tools.
- Implement Data Loss Prevention (DLP) controls to detect and block exfiltration of source code, PII, or confidential data to unapproved external platforms.
- Require organizational AI tool accounts to be provisioned and managed centrally through IT/Security rather than ad hoc by individual users.
Detection measures
- Monitor for employee access to new or unrecognized AI platform organizations via browser activity or network proxy logs.
- Establish anomaly alerts for bulk transfers of sensitive file types (e.g., `.py`, `.env`, `.sql`) to cloud-based AI services.
- Conduct periodic phishing simulation exercises that include AI platform impersonation scenarios to gauge employee susceptibility.