Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Fake banking sites aren’t just a customer problem. CFPB guidance makes clear that when a fraudster obtains account access information through deception and uses it to initiate a covered EFT, the transfer may qualify as an unauthorized EFT under Regulation E. That means cloned login pages can create investigation obligations, provisional credit requirements, and reimbursement exposure for banks, even when the customer typed the password themselves.

Article by Luis Monzon originally published on MyBroadband, May 9, 2026. Anthropic’s AI model Mythos, part of its Claude software, represented a clear threat to banks and financial institutions in South Africa. This is according to Bijan Sanii, CEO of Canadian fraud detection provider INETCO. Anthropic, one of the world’s most important generative AI makers, positioned Mythos as an extremely capable AI model designed to identify vulnerabilities in critical software beyond human capabilities.

ATO fraud cost US adults $15.6 billion in 2024, yet most fraud teams are still measuring detection time from the moment an alert fires, not from the moment an attacker starts building infrastructure. That gap is where the damage happens. To reduce time to detect fraud, teams need to move detection upstream, to Stage 1 and Stage 2 of the fraud lifecycle, before phishing sites go live and before a single credential is submitted. Faster transaction monitoring won’t close this gap.

A recent American Banker article, “Knock on wood: Are banks doing enough to cope with Mythos?” raises a timely and uncomfortable question about advanced AI models like Anthropic’s Claude Mythos. As highlighted in the article, INETCO CEO Bijan Sanii points out a critical truth: The conversation is being fueled by the emergence of AI technology capable of identifying software vulnerabilities at a speed and scale that was previously unimaginable.

Most enterprise fraud stacks are built to detect account takeover after it’s already succeeded. Login anomaly rules fire at authentication. Transaction models fire at monetization. By both points, the attacker is already inside. Knowing how to detect account takeover in real-time means shifting detection upstream – to behavioral signals, device trust, credential exposure feeds, and session integrity monitoring that activate before any fraudulent transaction is attempted.

Researchers at Google’s Threat Intelligence Group (GTIG) are tracking a new threat actor that’s impersonating help desks to trick users into installing malware. The threat actor, which GTIG tracks as “UNC6692,” begins by sending a large volume of spam emails to the victim, then initiates contact via Microsoft Teams to ostensibly help the user block the spam.

Most brand impersonation protection programs are built around a process that starts after the damage is done. A fake site goes live. Customers land on it. Credentials get stolen. Then the takedown request goes in. That sequence isn’t a workflow problem. It’s an architectural one. Preemptive brand impersonation protection means intervening before credentials are entered, not after a cloned site is discovered.

Deepfake-driven fraud has caused $2.19 billion in losses globally, with $1.65 billion reported in 2025 alone, according to an analysis by Surfshark. More than half of these losses were due to investment scams using deepfakes of high-profile figures.

Phishing reports and customer complaints are not early warning signals. By the time they arrive, attackers have already built the infrastructure. Lookalike domains are live, credential harvesting pages are indexed, and the exposure window is open. To stop digital impersonation attacks, organizations need to shift detection to the infrastructure preparation stage, before distribution begins.