As reported by SecurityAffairs, SOCRadar's analysis of stealer logs from the past 90 days has uncovered a worrying trend: corporate AI accounts are now a primary target for infostealer malware, with 482 companies showing exposed credentials and 295 of those appearing in actively circulating logs. The headline figure — ChatGPT/OpenAI sessions appearing at roughly 90% of affected organizations — tells us less about OpenAI's security posture and more about the scale of unsanctioned AI adoption in the enterprise.

AI Security Alert: As reported by SecurityAffairs, SOCRadar's analysis of stealer logs from the past 90 days has uncovered a worrying trend: corporate AI accounts are now a primary target for infostealer malware, with 482 companies showing exposed credentials and 295 of those appearing in actively circulating logs.

Session Theft Is Not Credential Theft — and Most Teams Still Don't Get This

The most critical insight from this report is one that many security programs still fail to operationalize: a stolen session cookie or API key is a live authentication artifact. Rotating the user's password does nothing to invalidate it. The attacker remains logged in, often indefinitely, until the session expires or is explicitly revoked. Okta's Jeremy Kirk, cited in the report, demonstrated this after pulling a 7 GB stealer dump from Telegram containing thousands of still-replayable tokens — including two dozen valid API keys for major AI providers.

The standard incident response playbook for credential compromise — reset the password, enforce MFA, move on — is fundamentally inadequate when the stolen artifact is a session token. You are closing the front door while the attacker is already sitting in the living room.

Why AI Accounts Are a Categorically Different Exposure

A compromised AI account is not equivalent to a compromised email or CRM login. It is arguably worse, for three reasons:

Why AI Accounts Are a Categorically Different Exposure
Conversation history is proprietary data. Employees routinely paste source code, internal documents, strategic plans, and customer data into AI prompts. That history persists in the account. An attacker with session access can read it all without triggering any additional authentication.
API keys enable compute theft and downstream compromise. Stolen API keys can be used to run inference workloads at the victim's expense, but more dangerously, they can be leveraged to access connected integrations, data pipelines, and third-party services that trust the AI platform's identity.
Shadow AI means no visibility. Many of these accounts were created with work emails on personal devices, outside IT oversight. Security teams don't know these accounts exist, so they can't monitor for compromise, enforce session policies, or respond when tokens appear in stealer logs.

The Claude Absence Is a Warning, Not a Reassurance

SOCDar's finding that Claude and Gemini barely appear in the data should not be interpreted as evidence of superior security on those platforms. As the report notes — and as Anthropic's own August incident confirmed — attackers hijacked Claude sessions to drain paid usage credits, forcing company-wide sign-outs and fraud refunds. The low volume simply reflects lower corporate adoption volume today. As usage grows, so will the targeting. Defenders who treat this as a ChatGPT-specific problem are making a dangerous assumption.

Shield53 Recommendations

  • Implement session revocation in your IR playbook. When an AI account compromise is suspected, the response must include forcing global session invalidation through the provider's admin console — not just a password reset. Document the specific steps for each AI platform your organization uses.
  • Discover and inventory shadow AI. Use CASB, proxy logs, DNS analysis, or SaaS discovery tools to identify all AI services accessed from corporate identities. You cannot protect what you have not found.
  • Rotate and scope API keys aggressively. Enforce least-privilege scoping on all AI API keys, set expiration policies, and integrate key rotation into your secrets management pipeline. Treat AI API keys with the same rigor as cloud provider credentials.
  • Monitor stealer logs proactively. Subscribe to threat intelligence feeds that surface your domains and identities in infostealer dumps. Time-to-detection matters — a session token found and revoked within hours is far less damaging than one discovered weeks later.
  • Deploy endpoint detection for infostealer families. The root cause here is not the AI platform — it is infostealer malware on employee devices. Ensure your EDR coverage includes detection for prevalent stealer families like Lumma, RedLine, Stealc, and Raccoon.
  • Establish an acceptable AI use policy. Provide sanctioned AI tools with enterprise governance rather than forcing employees into unsanctioned personal accounts. Banning AI use without providing alternatives drives shadow adoption.

The broader implication is clear: as AI platforms become embedded in daily work, the session tokens and API keys associated with them are now first-class attack targets. Organizations that continue to treat AI account compromise as a routine credential issue will find themselves responding to incidents that their playbook was never designed to handle.