A recent report indicates that threat actors are leveraging information stealer logs to obtain replayable AI tokens, which can then be used to bypass multi-factor authentication (MFA) and gain unauthorized access to AI user accounts. These "stolen keys" are reportedly being used to access tools provided by major AI model providers, including Google and Anthropic.
The mechanism involves common information stealer malware, such as Lumma Stealer and Vidar, which are designed to exfiltrate a broad spectrum of sensitive data from compromised systems. This harvested data typically includes traditional credentials, but critically, also encompasses session tokens and API keys. It is these session tokens and API keys, specifically those associated with AI services, that are being identified as replayable and capable of bypassing MFA.
When a user interacts with an AI service, a session token or API key is often generated to maintain authenticated access without requiring repeated credential entry. If an information stealer compromises a system, it can harvest these active tokens. The replayability of these tokens means that a threat actor can use them from a different location or device to impersonate the legitimate user, effectively bypassing any MFA prompts that would normally challenge a new login attempt.
This class of attack highlights a persistent challenge in cybersecurity: the protection of active session material. While MFA significantly strengthens initial authentication, its effectiveness can be diminished if active session tokens are compromised and are not sufficiently bound to specific user environments or invalidated upon suspicious activity. Products in the AI service category, like many cloud-based platforms, rely on these tokens for seamless user experience, making their secure handling paramount.
The scope of this issue is potentially broad, given the widespread use of information stealer malware and the increasing adoption of AI tools across various industries. Any organization or individual using AI services from providers like Google and Anthropic, whose systems might be susceptible to information stealer infections, could be at risk.
Mitigation strategies for this type of threat typically involve a multi-layered approach. Users should employ robust endpoint security solutions to prevent information stealer infections, practice good cyber hygiene, and be wary of phishing attempts. For organizations, implementing strong access controls, monitoring for anomalous login patterns, and ensuring that session tokens have appropriate expiration times and are invalidated upon suspicious activity are crucial. Additionally, where possible, binding session tokens to specific device identifiers or IP addresses can reduce their replayability.
This incident underscores the evolving threat landscape where traditional credential theft is now augmented by the compromise of session and API tokens, particularly in the context of emerging technologies like artificial intelligence. It serves as a reminder that security measures must continuously adapt to protect not just initial authentication, but also the ongoing authenticated sessions that power modern digital services.
