arXiv: PhiShark2026: A Multi-Layer Active-Web Raw-Evidence Dataset for Phishing Website Research

A new research paper, PhiShark2026, has been published on arXiv, introducing a large-scale dataset of phishing websites designed to improve detection systems. The dataset is notable for its multi-layer approach, capturing raw evidence from active web pages, including visual, textual, and structural elements, rather than relying solely on static URLs or blacklists. This is a research contribution, not a new regulation, but it signals a shift in how phishing threats are analyzed and mitigated.

The primary audience is cybersecurity vendors, financial institutions, and any organization operating online services that face phishing attacks. Compliance teams in banking, e-commerce, and critical infrastructure should pay attention because this dataset could lead to more sophisticated phishing detection tools, which in turn affect how you assess third-party risk and incident response. Regulators may also reference such datasets when evaluating the adequacy of security controls under frameworks like DORA or NIS2.

Compliance teams should monitor this development and assess whether their current anti-phishing controls are capable of handling the type of raw, multi-layered evidence this dataset represents. You should review your vendor due diligence to see if your security providers are incorporating such advanced datasets into their products. Finally, update your internal threat modeling and incident response playbooks to account for more complex phishing attacks that bypass traditional URL-based filters, and consider how your evidence collection procedures would hold up if a regulator asked for proof of detection capability against these newer attack vectors.