AI regulation is getting concrete. Here’s what it asks for.
Nov 17, 2026
AI Regulatory Readiness series, Part 1 of six6Different AI laws and frameworks keep asking for the same security controls: access, integrity, monitoring, and incident reporting. Build around those controls rather than one regulation at a time, and you can extend the identity, access, and data security you already run.
For most of the last few years, AI governance was a conversation about values like transparency, fairness, and accountability. Those principles mattered, but they were hard to hand to a security team. You can’t configure fairness on a file server.
But that’s changing. A new wave of AI laws and security frameworks is translating broad principles into operational expectations: who can access AI systems, how their data is protected, how integrity is maintained, and how incidents are detected and reported. In other words, AI regulation is turning security controls into business requirements.
This series breaks down what that shift means for security and compliance leaders, and how to prepare for it without needing to rebuild your security program from scratch.
There is no single AI compliance checklist
AI regulation isn’t one thing. Different laws apply to different organizations, systems, and use cases. A company training frontier-scale models faces very different obligations than an enterprise using AI to make automated decisions about customers. And voluntary frameworks sit alongside enforceable statutes, each carrying its own weight.
Here’s a simplified view of the landscape security leaders are navigating:
Source | Primary focus | Why security teams should care |
|---|---|---|
|
California SB 53, New York RAISE Act, Illinois AISMA |
Frontier AI developers |
Protecting unreleased model weights, incident visibility, and reporting |
|
EU AI Act |
Risk-based AI regulation |
Cybersecurity resilience for high-risk systems, including confidentiality attacks, data poisoning, and model poisoning |
|
Texas TRAIGA, Colorado ADMT |
AI use, data, and consumer protection |
Personal data security, data documentation, and selected governance requirements |
|
NIST AI RMF |
Voluntary AI risk management |
Security and resilience, including unauthorized access, exfiltration, and data poisoning |
|
OWASP LLM Top 10, MITRE ATLAS |
Technical AI threats and mitigations |
Sensitive data exposure, excessive agency, credential abuse, exfiltration, and poisoning |
|
ISO/IEC 42001 |
AI management systems |
Governance, lifecycle, data, monitoring, and management system controls |
Read row by row, the table looks fragmented: state laws aimed at a handful of AI developers, a sweeping EU regulation, consumer protection rules, voluntary frameworks, threat catalogs, and a management system standard.
But read down the right-hand column and the same words keep appearing: access, exfiltration, poisoning, integrity, monitoring, and incident reporting.
From principles to controls
Lawmakers and standards bodies are working on their own timelines, for different audiences, with different enforcement models. Yet they are converging on a recognizable set of security expectations.
Instead of asking “which AI law applies to us, and what does it say?” for every new statute, security leaders should ask a more durable question: “which security controls keep showing up, and do we have them?”
The regulatory map will keep moving. New state laws will pass, frameworks will be revised, and AI architectures will evolve from chat assistants to autonomous agents. Building clause by clause around one regulation means rebuilding each time the rules change. A program built around the recurring controls will adapt.
AI security isn’t separate from enterprise security
AI systems don’t run in isolation. They rely on the same identities, permissions, data stores, endpoints, and infrastructure your organization already manages.
A model’s weights sit in a repository someone has access to. Training data lives on file shares and in data platforms with permissions attached. AI assistants act through user and service accounts, agents use credentials, and all of it runs on endpoints and servers your team already secures.
Strengthening the controls you already operate around identity, access, and data is a practical path toward AI regulatory readiness.
Two requirements keep showing up
Two technical requirements run through the laws and frameworks:
- Protect AI models and related assets from unauthorized access or exfiltration. Model weights and AI artifacts are valuable intellectual property, and for some developers they are now an explicit regulatory concern.
- Protect training data and model logic from unauthorized modification or poisoning. If someone can change the data a model learns from, or the model itself, they can change how it behaves.
One is about theft, the other is about tampering. In Part 2, we’ll look at each in detail, and at why the same security foundation addresses both.
Next in the series: Part 2, Theft and tampering: The two risks behind AI security requirements.
Want the full picture now?
Download the AI Regulatory Readiness eBook for the complete landscape, the six core controls, and a five-step roadmap.
Download nowIn this series
- AI regulation is getting concrete. Here’s what it asks for.
- Theft and tampering: The two risks behind AI security requirements
- How AI amplifies the security weaknesses you already have
- Six controls that outlast any single AI regulation
- Securing AI takes a lifecycle approach
- A five-step roadmap to AI regulatory readiness
Share on
Learn More
About the author
Istvan Molnar
IT Security Compliance Specialist and Product Marketing Manager
Istvan Molnar is an experienced IT Security Compliance Specialist and Product Marketing Manager at Netwrix, with over a decade of expertise in international standards, regulations, and cybersecurity frameworks. He specializes in bridging the gap between complex compliance requirements and the Netwrix product portfolio, offering strategic guidance, compelling content, and support for compliance-driven initiatives and go-to-market strategies.
Learn more on this subject
The riskiest identities aren’t people: A partner guide to non-human identity governance
How AI amplifies the security weaknesses you already have
Theft and tampering: The two risks behind AI security requirements
From reactive auditing to proactive governance: The importance of a true data security platform
What is data exfiltration? How AI speeds up data theft and how to reduce what's exposed