What does the NYDFS AI cybersecurity guidance require of covered entities?

The NYDFS AI cybersecurity guidance is a pair of industry letters from the New York State Department of Financial Services telling DFS-regulated banks, insurers and other licensees how to read AI-driven threats into the cybersecurity program they already run under 23 NYCRR Part 500. The first letter went to the executives and information security personnel at all entities regulated by the New York State Department of Financial Services on October 16, 2024, and a second letter on frontier AI models followed on May 21, 2026.

What is the NYDFS AI cybersecurity guidance?

It is guidance explaining how the existing Part 500 Cybersecurity Regulation applies to AI, not a standalone AI rulebook. The 2024 letter states plainly that the guidance does not impose any new requirements beyond obligations already in DFS's cybersecurity regulation codified at 23 NYCRR Part 500.

Part 500 itself remains the binding obligation for covered entities, a New York licensed bank, insurer, or other DFS-covered entity, and it is what examiners actually enforce.

What cybersecurity risks does NYDFS tie to AI?

NYDFS names four risk areas, two created by threat actors' use of AI and two created by a covered entity's own reliance on AI. A law firm summary of the letter describes these as AI-enabled social engineering, AI-enhanced cybersecurity attacks, exposure or theft of nonpublic information, and third-party or vendor dependency risk.

On social engineering, the letter warns that threat actors are increasingly using AI to create realistic and interactive audio, video, and text deepfakes that allow them to target specific individuals via email, telephone, text, videoconferencing, and online postings.

On AI-enhanced attacks, it notes that because AI can scan and analyze vast amounts of information much faster than humans, threat actors can use AI quickly and efficiently to identify and exploit security vulnerabilities, often allowing them to access more information systems at a faster rate.

On exposure of data, it points out that products that use AI typically require the collection and processing of substantial amounts of data, often including nonpublic information.

On supply chain risk, each link in a vendor chain introduces potential security vulnerabilities that can be exploited by threat actors.

What does Part 500 require covered entities to do about those risks?

The 2024 letter directs covered entities to address these AI-related threats within the risk assessments and controls Part 500 already mandates, rather than creating a separate control family. It directs covered entities to address AI-related threats, including AI-enabled social engineering and deepfakes, AI-enhanced attacks, theft of nonpublic information used to train AI, and third-party and vendor AI risk, within their existing Part 500 risk assessments and controls.

Concretely, Part 500 requires covered entities to maintain cybersecurity programs, policies, and procedures that are based on cybersecurity risk assessments, and the guidance says those assessments must now take AI into account.

The regulation also requires covered entities to implement multi-factor authentication and other access controls in case MFA fails, which the guidance says should resist AI-enhanced social engineering.

Separately, the Cybersecurity Regulation requires covered entities to provide cybersecurity awareness training for all personnel, including specific training on social engineering, which the guidance says should now cover deepfake scenarios.

It also requires covered entities to implement monitoring processes to identify new security vulnerabilities and to monitor the activity of users and web traffic to protect against malicious activity, extended by the guidance to cover AI application behavior.

How does the 2026 frontier-AI advisory expand on this?

In a May 21, 2026 letter, NYDFS warned that certain frontier artificial intelligence models amplify the potency, scale, and speed of identifying vulnerabilities and exploits in information systems. The Department is issuing this Advisory about the heightened cybersecurity risks associated with certain frontier artificial intelligence models that amplify the potency, scale, and speed of identifying vulnerabilities and exploits in information systems.

It advises regulated entities to reassess their procedures for evaluating the criticality and threat of known vulnerabilities and to review vulnerability management timelines to determine whether accelerated detection and remediation processes are necessary based on updated risk assessments.

It also tells regulated entities to develop and maintain dependency maps, and coordinate with critical third-party service providers and material downstream providers to address significant vulnerabilities and operational risks.

On AI-assisted coding, it recommends that organizations restrict and validate inputs prior to running scripts or processes and confirm that secure programming practices are used.

On alerting, it recommends that entities ensure that suspicious activity is promptly flagged and addressed.

Like the 2024 letter, the 2026 advisory states it does not impose any new requirements for regulated entities; it is intended to inform their risk management and compliance efforts.

Is this guidance separate from federal model-risk rules?

Yes, Part 500 is described as a cybersecurity overlay that sits apart from federal model-risk guidance for banks. A comparison of the two regimes notes that SR 26-2, issued in April 2026, is federal model-risk guidance that replaced SR 11-7 and excludes generative and agentic AI from its scope.

A covered bank can sit under both regimes at once, and the two ask different questions about a given AI system.

What should security leaders do?

The practical task is a readiness exercise, not a new program: map each AI risk NYDFS named to the Part 500 control it touches and assemble the evidence an examiner would expect. Security leaders should update their annual Part 500 risk assessment to name AI-enabled social engineering, AI-enhanced attacks, NPI exposure from AI tools, and AI vendor risk as explicit threat scenarios, since the Cybersecurity Regulation requires Risk Assessments to be updated at least annually and whenever a change in the business or technology causes a material change to a Covered Entity's cybersecurity risk.

They should also extend third-party due diligence to AI vendors, retire overly permissive access methods, and test whether logging and alerting would catch AI-accelerated exploitation of a known vulnerability before it spreads. Because NYDFS frames both letters as interpretation of Part 500 rather than new rules, the deliverable for a board or examiner is documentation showing the existing cybersecurity program already accounts for AI as a threat, not a parallel AI compliance project.