Credential Malware Expands Into Enterprise AI Identity Theft
Information-stealing malware is increasingly harvesting corporate AI credentials, active sessions and API keys, creating a new route into sensitive business data and expensive computing resources.
Recent SOCRadar research cited by Security Affairs analyzed stealer-log activity from a 90-day period and identified 482 companies with exposed AI-related accounts or credentials.
Of those organizations, 295 appeared in active logs during the period. Researchers associated the exposure with 5,434 stealer-log records and roughly 1,500 distinct corporate email addresses.
ChatGPT Dominates Observed Exposure
ChatGPT and OpenAI-related sessions represented the largest concentration in the dataset.
358 Companies Had OpenAI-Related Exposure
Researchers identified captured ChatGPT or OpenAI sessions associated with 358 of the 482 affected companies, accounting for roughly 90% of the records analyzed.
Platforms including Zapier, Notion, Hugging Face, Replit, Lovable and ElevenLabs also appeared.
The high ChatGPT figure should not be interpreted as evidence of a vulnerability in OpenAI's platform. It more likely reflects widespread workplace adoption—including employees using corporate identities from personal or unmanaged devices.
Session Theft Can Defeat a Password Reset
The most important technical distinction involves what an infostealer actually captures.
Cookies Can Preserve Authenticated Access
Infostealers frequently collect more than usernames and passwords. They can extract browser cookies, authentication tokens and active session material.
If an attacker can replay a valid session token, changing the victim's password may not immediately terminate the stolen session.
That makes session revocation a critical part of incident response rather than relying solely on credential rotation.
AI Conversations Become Sensitive Corporate Data
Compromising an AI account can expose information that never existed inside a conventional corporate database.
Prompt Histories May Contain Internal Secrets
Employees increasingly place source code, contracts, customer information, technical documentation and business plans into generative-AI conversations.
If conversation history remains accessible through a stolen authenticated session, the attacker may obtain that information without directly breaching the organization's internal network.
AI accounts are therefore becoming another repository of corporate memory that security teams must protect.
API Theft Creates LLMjacking Opportunities
AI credentials also have direct monetary value.
Attackers Can Consume the Victim's Compute Budget
Stolen API keys can be abused to run AI workloads against the victim's paid account—a practice commonly described as LLMjacking.
Attackers can consume model quotas themselves or resell unauthorized access through underground markets.
Google Threat Intelligence reported increased buying and selling of AI accounts during 2026, with demand particularly visible for Claude and Gemini credentials as well as AI development tools such as Cursor Pro and Devin.
Stealers Are Targeting Developer Configurations
The threat is already moving beyond browser sessions.
Google researchers observed infostealer operators targeting configuration stores belonging to AI coding assistants, including files capable of containing plaintext API keys and custom model endpoints.
Successful theft could expose both paid model quotas and infrastructure connected to developer workflows.
Connected AI Services Increase Lateral Risk
The consequences become more serious when compromised accounts are integrated with other enterprise platforms.
Automation Tokens Can Extend the Breach
A stolen automation-platform session may carry existing authorization into email, CRM platforms, cloud storage or other SaaS applications.
This changes the security model considerably: compromising one AI or automation identity can potentially provide access to systems that trust that identity downstream.
Shadow AI Is Becoming an Identity-Security Problem
The underlying weakness is often visibility rather than the AI platform itself.
Employees can create AI accounts using corporate email addresses from devices that are outside enterprise management, leaving security teams unaware that those identities exist.
Organizations should therefore treat AI services as part of their identity attack surface: discover unsanctioned accounts, move approved platforms behind SSO where possible, shorten session lifetimes, rotate API credentials, revoke active sessions after compromise and investigate the endpoint responsible for any stealer-log exposure.
The larger security shift is clear: AI accounts are no longer merely productivity-tool logins. They can contain sensitive conversations, valuable compute entitlements and trusted connections to other business systems—making them increasingly attractive assets in the infostealer economy.