536/69 Wednesday, September 30, 2026
SOCRadar has released an analysis of stealer log data collected over the past 90 days, identifying 482 organizations with exposed AI accounts or credentials. Of these, 295 organizations appeared in logs that remained active during the same period, indicating that many of the risks were linked to recent activity rather than older data collected after the fact. In total, researchers identified 5,434 stealer log records associated with 1,500 corporate email addresses. ChatGPT and OpenAI accounts or sessions appeared across 358 of the 482 affected organizations, accounting for approximately 90% of the records covered in the report, while other platforms such as Zapier, Notion, Hugging Face, Replit, Lovable, and ElevenLabs appeared at significantly lower rates.
The report emphasized that these figures do not indicate a direct security issue with OpenAI, but instead reflect how AI services are being used in workplace environments. Many employees register for ChatGPT using corporate email addresses on personal devices or use AI services outside organizational policies without the knowledge of IT teams. A key concern is that stolen AI accounts may expose not only passwords, but also session cookies and API keys, which attackers may replay to bypass authentication. Even after a password is changed, an attacker may remain logged in if an existing session has not been revoked. In addition, AI conversation histories may contain source code, customer data, contracts, unreleased plans, or other internal information previously entered by employees, effectively turning AI accounts into part of an organization’s “institutional memory” that attackers can access without directly breaching internal systems.
Beyond data exposure, stolen API keys can also create financial risk through activity known as LLMjacking, where attackers obtain keys from notes, configuration files, or workspace settings and use them to run AI workloads at the victim’s expense or resell access at a low price. Accounts on automation platforms such as Zapier can introduce additional risk because they may be connected to CRM systems, email, and file storage, allowing attackers to create workflows that continuously exfiltrate data through infrastructure belonging to trusted service providers. The report recommends that organizations integrate AI platforms with Single Sign-On, use short-lived sessions, regularly rotate refresh tokens and API keys, and detect access from unusual locations or at abnormal times. If an employee appears in a stealer log, organizations should treat the incident as an endpoint security incident rather than simply resetting the password. They should also identify shadow AI accounts before they can effectively control and reduce the associated risk.
Source: https://securityaffairs.com/199933/ai/ai-accounts-are-becoming-the-new-target-for-infostealers.html