The NYDFS industry letter of October 16, 2024 does not add new rules. It explains how existing Part 500 duties apply to AI risk: include AI in your risk assessment, vet vendors that use AI, strengthen access controls, train staff, monitor for misuse and keep an inventory of AI systems. Engineering teams can turn that into a short checklist.
This is general information, not legal advice.
What the letter is
The Department says the guidance explains how covered entities should use the framework in Part 500 to assess and address cybersecurity risks arising from AI. It names four risks: AI enabled social engineering, AI enhanced attacks, exposure or theft of large amounts of nonpublic information, and added vulnerabilities from third party and supply chain dependencies. The first two come from attackers' use of AI. The last two come from your own use of or reliance on AI.
Checklist for engineering
- Risk assessment. The letter says assessments should address the organization's own use of AI, the AI used by vendors, and vulnerabilities in AI applications. Add AI tools used by developers to the list.
- Inventory. If you use AI or rely on a product that does, the letter says to identify all information systems that use or rely on AI and keep an inventory, prioritizing mitigations for critical ones.
- Data discipline. Staff who use AI applications should be trained to draft queries that avoid disclosing nonpublic information. The letter also suggests considering monitoring for unusual query behavior and blocking queries that could expose nonpublic information to a public AI product.
- Access controls. The letter encourages authentication factors that resist deepfakes and suggests avoiding SMS, voice or video authentication in favor of digital certificates and physical security keys.
- Training. At least annual awareness training that includes social engineering, and, where AI is deployed, training on securing and developing AI systems.
- Vendor terms. See the next section.
Vendor diligence for AI
The Department strongly recommends that third party service provider diligence consider the threats vendors face from AI and AI enabled products, how those could affect you, and how vendors protect themselves. It suggests contract terms for timely notice of cybersecurity events and, where a vendor uses AI, additional representations and warranties about secure use of your nonpublic information.
Data minimization
The letter ties AI to data management. Part 500 requires disposal of nonpublic information no longer needed, including data used for AI purposes. Less retained data means less to lose.
A small first step
Write a one page AI use standard that covers approved tools, what data may never be entered, who approves new tools, and where the inventory lives. Review it with your risk assessment each year.
Mistakes to avoid
Teams stumble when approved AI tools are listed but personal accounts are not addressed, when the training covers social engineering but not what data to keep out of prompts, and when the inventory counts only products the company bought and misses AI features switched on inside existing tools. Ask for a quarterly check of enabled AI features in your main developer and business tools, and keep the list current.
Quick check
Is the letter a new regulation?
No. It is guidance on applying existing Part 500 requirements to AI related risk.
Does it ban public AI tools?
No. It suggests considering monitoring and blocking queries that might expose nonpublic information to a public AI product, as a risk based choice.
Sources
Last reviewed: 19 October 2026.
