Cequence Security Uncovers Hackers Impersonating ChatGPT and Claude to Steal Banking Credentials

Cequence Security has identified a sophisticated attack vector where malicious actors impersonate over 30 AI bots, including ChatGPT and Claude, to infiltrate financial networks. This tactic bypasses traditional IP-based security to harvest sensitive credentials, highlighting a critical vulnerability as banks begin to integrate legitimate AI agent traffic into their digital ecosystems.

What was announced

Cequence Security researcher Will Glazier has uncovered a campaign where a single attacker cycled through approximately 32 different AI bot identities, including OpenAI’s ChatGPT, Anthropic’s Claude, Meta, and Perplexity. On a single day this month, the attacker targeted nearly half of the companies protected by Cequence that receive AI bot traffic, with financial institutions identified as primary targets. The scanner utilized more than 300 cloud addresses in a single day, a tactic designed to render traditional IP-based blocking ineffective. Unlike legitimate AI crawlers, this bot specifically hunts for sensitive configuration data, such as .env files, cloud credential files, and private SSH keys.

The research notes that while legitimate AI agent-driven banking activity is currently minimal—representing roughly one in every 4,000 requests even at highly exposed banks—the rapid growth of these services provides a convenient mask for cybercriminals. The attacker was observed switching identities within seconds from the same machine, claiming to be different trusted agents to bypass security perimeters. To counter these threats, the research suggests that organizations should verify AI crawlers against published address lists or signed requests rather than relying on user-agent headers. Additionally, it recommends that banks admit third-party agents only through registered integrations on API gateways, ensuring access tokens are tied to specific customers and tasks.

"Banks are currently scrambling to figure out which AI bots belong on their networks, but cybercriminals have already moved on to the next phase. They are no longer trying to break into these systems. Instead, they are simply logging in disguised as trusted AI agents."

Shane Barney, Chief Information Security Officer at Keeper Security

The companies involved

Cequence Security, reachable at cequence.ai, is a cybersecurity firm focused on protecting APIs and web applications from automated threats. Based in Sunnyvale, California, the company provides a platform designed to identify and mitigate sophisticated bot activity that targets sensitive data and business logic. The research also features insights from leaders at DataVisor and Keeper Security. Patrick Harr serves as the Chief Executive Officer at DataVisor, an AI-powered fraud and risk management platform that specializes in detecting coordinated attacks and financial crime in real-time. The firm focuses on using machine learning to stay ahead of polymorphic fraud tactics that evolve faster than legacy systems. Shane Barney is the Chief Information Security Officer at Keeper Security, a provider of zero-trust and zero-knowledge cybersecurity software, primarily known for its password management and privileged access management solutions. The company emphasizes the protection of digital assets through encrypted vaults and secure credential sharing. Together, these organizations represent a cross-section of the security industry addressing the intersection of artificial intelligence, credential theft, and the shifting landscape of financial services infrastructure.

What this means

This discovery signals a shift in the cat-and-mouse game between financial institutions and attackers. As banks race to implement AI-driven customer experiences, they are inadvertently expanding their attack surface. The fact that attackers are already successfully masquerading as trusted agents like ChatGPT suggests that the "honeymoon period" for AI integration is over before it truly began. Legacy security models that rely on IP reputation or simple user-agent strings are now effectively obsolete against polymorphic AI threats. The industry is under immediate pressure to move toward cryptographic verification and rigid API gateway controls. The open question is whether financial institutions can balance the friction of these measures with the seamless experience promised by AI agents.