Infostealer Steals AI Cookies, Bypasses Two

Infostealer, a type of malware specializing in stealing user passwords, credit card information, and cryptocurrency wallet data, has now set its sights on generative AI accounts. Infostealer infiltrates PCs and browsers to extract stored passwords, login credentials, and financial data. As sensitive information such as work documents and source code accumulates in AI platforms, hacking groups are increasingly targeting AI accounts.

According to a September 18 report by the Nikkei Shimbun, Israeli cybersecurity firm KELA confirmed that over 49,700 cases of session cookies (authentication data) stolen from AI platforms were circulated on the dark web from the beginning of this year until last July. A significant portion of these cookies remained valid for actual logins. Session cookies serve as proof that a user has completed login procedures, including passwords and two-factor authentication. Stealing them allows attackers to access AI accounts without needing passwords, effectively bypassing the "front door" of two-factor authentication by stealing the user’s "access pass."

◇Stealing Personal Information and Corporate AI Computing Resources in One Go

AI accounts have become new targets for cybercrime because a single account can yield both "information" and "computing resources." General users often personal details like interests, concerns, schedules, or upload documents and photos for analysis via AI. For businesses, the risks are even greater: employees use AI to summarize internal documents, analyze business strategies or financial data, and developers input source code to detect errors or generate code.

In March, Microsoft investigated malicious browser extensions that secretly collected ChatGPT and DeepSeek conversation data. It found approximately 900,000 cases of installations and detected activity linked to these extensions in over 20,000 corporate or institutional work accounts. Microsoft noted that AI conversations could contain sensitive information such as source code, internal procedures, and strategic discussions.

With rising AI computation costs, "access to AI services" has also become a target. Stealing a company’s API (application programming interface) keys allows criminals to exploit high-performance AI at the victim’s expense. Octa, a U.S. authentication and security firm, analyzed 7GB of Infostealer data and found information extracted from 5,871 infected PCs across 162 countries. Among 44,791 authentication tokens (used to verify login status and user permissions), 555 were linked to AI services. API keys for platforms like Google Gemini and OpenAI were also discovered.

There are confirmed cases of stolen API keys being misused. Anthropic recently revealed that hackers stole AI API keys from companies and used them to attack other businesses at the victims’ expense. A criminal ecosystem is emerging where stolen keys and accounts are sold on dark markets or supplied to illegal AI services. Anthropic warned, "Stealing AI authentication data gives criminals three advantages: resalable assets, computational resources at others’ expense, and the ability to impersonate victims."

◇AI as a Gateway to Attack Other Systems

As AI evolves from simple chatbots to agents performing human tasks, the damage from such breaches is expected to grow. AI coding tools and agents connect to external systems like GitHub, databases, and clouds based on their settings. During this process, authentication data such as API keys and access tokens for other systems may be stored on PCs.

A source from the tech industry stated, "When AI was merely a tool for answering questions, account theft was limited to stealing conversation history or access rights. But as AI agents now connect to multiple work systems, a single compromised account could trigger a chain of security incidents." They added, "The more tasks AI can perform, the more damage attackers can inflict through stolen accounts."