Stolen AI Tokens Bypass MFA in Infostealer Log Attacks

Cybercriminals are weaponizing information stealer logs to hijack accounts on leading artificial intelligence platforms, bypassing even multi-factor authentication through replayed session tokens and API keys. According to a new report from Okta's threat intelligence team, malware families like Lumma Stealer and Vidar are harvesting credentials, session tokens, and API keys from infected machines, then bundling them into stealer logs sold on underground forums. These logs contain reusable authentication artifacts that let attackers impersonate legitimate users without ever supplying a password or completing an MFA prompt.

Okta analyzed a 7 GB infostealer dump released on a Telegram channel on August 2, 2026, uncovering data from 5,871 infected machines across 162 countries. Within the dataset, researchers identified 44,791 unique JSON Web Tokens (JWTs), of which 555 were tied to authentication for AI services from Google, Microsoft, Anthropic, Amazon, Gamma, Notion, Character.ai, Cursor, Poe.com, and Pika AI. An additional 2,937 JSON Web Encryption (JWE) structures—mostly issued by OpenAI via NextAuth.js—were also present. "Session tokens and API keys are sought specifically by threat actors because it is often possible to replay those secrets and bypass credential-based authentication," explained Jeremy Kirk, Okta's director of threat intelligence. As long as these tokens remain unexpired, attackers can replay them to log directly into LLM services on day one.

The findings also expose a significant privacy risk: 17.7% of the 44,791 JWTs contained plaintext personally identifiable information (PII), including names, phone numbers, and email addresses. Unlike passwords, this PII does not expire and directly links users to specific services—prime material for targeted phishing or social engineering campaigns. On the day of the dump's release, 1,843 JWTs and JWEs were still active and exploitable, highlighting how token-based authentication creates a long-lived attack surface compared to traditional credentials.

To defend against session replay attacks, organizations should implement IP allowlisting to restrict authentication traffic to approved network ranges and shorten token lifetimes wherever possible. Individual users can take immediate action by checking whether their credentials have appeared in known breaches using an email breach checker, testing the strength of reused passwords with a password checker, and rotating any API keys tied to AI platforms immediately.