Today the world is celebrating artificial intelligence (AI) Appreciation Day and looking to recognize not only technological advancements that Generative AI (GenAI) has made, but also to discuss its future potential and legal implications. As the old joke goes, GenAI appears to be the “bacon” of business process as more and more companies think everything is better once you sprinkle GenAI on it. Of course, for those of us who are helping our clients implement these GenAI solutions, we know that while GenAI tools have tremendous power to help us understand our data, they also spin off unintended externalities within an organization that need to be addressed.

Indeed, it appears AI has moved beyond the innovation-only stage and entered into an era of accountable adoption. In Norton Rose Fulbright's 2026 Annual Litigation Trends Survey: A midyear industry pulse, nearly half of United States in-house counsel reported increased AI-related dispute exposure at both the federal (46 percent) and state (42 percent) levels. At the same time, 41 percent identified AI-enabled products and deployments as likely triggers of class action litigation this year. The practical message is not that organizations should slow innovation. It is that companies need to understand the key challenges and risks with AI adoption and consider how to strengthen AI governance controls so the benefits of AI can be realized defensibly.

AI regulations are proliferating rapidly, with a focus on high-risk systems and use cases

Although the US currently lacks a comprehensive federal AI law, we are tracking over 450 enacted and pending state laws regulating AI use. Notable among these are comprehensive AI governance laws like the Colorado AI Act and CCPA’s regulations governing automated decision making technology (ADMT) that are sector agnostic and may apply to a wide range of organizations. An emerging theme among these laws is that they impose different requirements on developers versus deployers of AI systems or ADMT and cover AI systems that make “consequential decisions” or “significant decisions” about a person particularly in high-consequence areas like employment, housing, healthcare, finance, insurance, education and legal services.

Outside the US, the EU AI Act remains the most prominent AI framework and continues to shape how many organizations think about risk-based AI governance even outside EU-regulated contexts. The EU AI Act is a risk-based regulation that applies more significant controls to higher-risk use cases, and we recently assessed draft guidance published in May of this year that further clarifies this risk hierarchy and how it should be applied for high risk systems.  On July 8, 2026, the EU enacted amendments to the EU AI Act to extend deadlines for compliance with high-risk system obligations, prohibit use of AI to general non-consensual intimate content and clarify certain definitions and requirements.

A common theme across these laws is that organizations remain responsible for how AI is used, even when they only deploy third party tools rather than develop them. Regulators expect companies to understand their AI use cases, conduct risk assessments, implement safeguards and monitor outcomes.

AI turbocharges traditional cybersecurity challenges

AI can increase cybersecurity threats by allowing threat actors to plan and execute attacks more efficiently and with more sophisticated techniques. It has been reported that attacks by Al-enabled adversaries increased 89 percent from 2025 to 2026, and as we have previously discussed, these threats appear likely to increase.

Organizations are also increasingly relying on AI, particularly third-party developed AI, to power business operations. This usage expands the attack surface because AI platforms and integrations can themselves be vulnerable to security and privacy issues. As companies adopt agentic AI systems that can take action without human intervention, we have also assessed the unique security challenges with agentic AI adoption.

Regulation and litigation focuses on AI chatbots and child protections

Recent AI legal developments also reflect the growing public and regulatory concern around child safety, mental health and consumer-facing AI systems, including so-called “companion chatbots.”  A growing number of states have proposed or enacted laws regulating AI chatbots and other consumer-facing AI tools to minimize risks to children and vulnerable populations. For example, New Hampshire’s AI Chatbot Child Safety Law creates criminal liability and a private right of action for AI chatbot or generative communication providers that knowingly direct communications to a child encouraging imminent harmful conduct. California requires companion chatbot operators to disclose their AI identity, implement self-harm prevention protocols and provide safeguards for minors, including periodic break reminders.

This use case is expected to be an active litigation area. An AI developer was recently sued over concerns that its popular AI chat product is purportedly addictive, harms minors and encourages vulnerable users toward suicide. In other words, AI is being treated as a product for product liability purposes.

Organizations using AI chatbots should consider how the chatbot is expected to be used and is actually being used; what notice and disclosures are appropriate; how user inputs will be processed and retained; and whether and how sensitive information may be shared through these platforms.

The absence of AI-specific regulations does not mean the absence of AI regulation

Existing legal frameworks already apply to the use of AI even if AI never appears in the text of the statute. For example, compare California which recently expanded its employment regulations to specifically address AI-assisted hiring and employment decisions, with the Equal Employment Opportunity Commission (EEOC) which identified "technology-related employment discrimination” as an enforcement priority, including the use of AI. Similarly, organizations may be navigating a number of new AI regulations targeting the healthcare industry like the Texas Responsible AI Governance Act (TRAIGA), but HIPAA will likely remain the technology-agnostic standard bearer for most healthcare privacy and security issues.

Regulators are also using existing laws to regulate AI even though those laws are not specific to AI.  For example, as we discussed previously, the Pennsylvania State Board of Medicine brought an action against a company under the existing Medical Practice Act in Pennsylvania alleging that AI platform may be liable under a state medical licensing statute for what a user-created chatbot character. The Pennsylvania action is a reminder that the risk landscape for conversational AI is not limited to AI-specific regulation or consumer protection law. Wherever an AI system's outputs make specific representations and venture into subjects generally reserved for licensed professionals, whether in medicine, law, mental health, or financial advice, companies should be aware that existing state licensing regimes may apply to those outputs and interactions.

Our take on best practices in AI governance

AI governance starts with understanding use cases

Perhaps the most important lesson from emerging AI regulation and litigation is that not all AI carries the same level of risk.  An internal chatbot helping employees locate policies presents very different legal and operational risks than an AI system used to screen job applicants, make insurance decisions, analyze healthcare information or interact directly with consumers. Emerging regulatory frameworks like the EU AI Act distinguish between these types of use cases and impose heightened obligations on higher-risk deployments, whereas other AI governance laws functionally recommend heightened obligations based on risk by requiring organizations to conduct risk assessments and implement appropriate controls accordingly.

An AI use inventory that sits on the shelf and gathers dust is not useful or compliance-minded. It should connect to a risk-based governance process that determines which uses are permitted, restricted, escalated or prohibited; what approvals are required before deployment; what contractual protections are needed for third-party tools; and what ongoing monitoring is required after a tool goes live and apply controls proportionate to identified risks.

Building (or maturing) an effective AI governance program

Ultimately, AI Appreciation Day is a reminder that innovation and governance are not competing objectives—they are complementary disciplines that help organizations realize AI’s potential while navigating an increasingly complex legal and regulatory landscape. AI governance programs are the means for implementing effective controls that will allow AI’s benefits to scale responsibly and sustainably.

To support responsible AI adoption while addressing the trends identified above, organizations should consider the following measures:

Develop clear policies and procedures for approved and prohibited uses

Organizations should identify which tools may be used, what data may be input, what outputs may be relied upon, when human review is required, when disclosure is required and which use cases are restricted or prohibited. An AI Use Policy may be a good place to start.

Establish cross-functional oversight with authority

AI governance programs typically work best when they include legal, privacy, cybersecurity, compliance, IT and business stakeholders. The governance function should have authority to approve, condition, pause or reject higher-risk deployments.

Create and maintain an AI inventory

This inventory should include approved tools, use cases, data categories, system owners and risk tiers.

Adopt a risk tiering framework

Low-risk internal productivity tools should not be governed the same way as higher risk AI systems. Risk tiers should drive approval requirements, legal review, technical testing, human oversight, monitoring and documentation.

Enhance and update internal data segmentation, data integrity and information governance

AI is as good as the data it gets and the more disorganized and dirtier the data, the less effective it is going to be. Unleashing AI in an environment with poor access controls can enable inadvertent (or worse) access to protected information by employees. In particular, legal departments risk waiving privilege if unnecessary employees access privileged documents through CoPilot or Gemini.

Build AI governance into existing workflows

AI review should be embedded into procurement, vendor onboarding, privacy impact assessments, cybersecurity reviews, product development, software development, contracting, records management, incident response and legal hold processes.

Use AI-specific vendor diligence and contract terms

Standard SaaS diligence may not be sufficient for AI tools. Vendor review should address model training, data use, confidentiality, retention, deletion, explainability, testing, cybersecurity, audit rights, regulatory cooperation, incident response, output ownership, among other topics.

Enhance employee AI literacy and consider training based on role

General AI literacy training is helpful, but higher-risk roles may need tailored training aligned with the risks of the tools they are using and information they are handling.

Organizations should consider whether their incident response plans, employee awareness, tabletop exercises and other measures designed to strengthen the information security program, account for AI-specific scenarios, including data leakage through prompts, supply chain attacks and vendor outages and agentic AI actions that exceed approved authority.

Monitor and refresh governance over time

AI governance programs should evolve as AI capabilities evolve, laws develop, employees adopt new workflows and organizations discover new uses. Periodic reassessments, audits and governance reporting can help ensure that controls remain aligned with actual use.

Additional resources

As the AI landscape evolves, there have been countless legal developments that we have not been able to cover here. Organizations looking to learn more may find the following 2026 insights helpful:

Federal vs. state AI regulation

A new White House national AI policy framework and DOJ’s recent challenge of the Colorado AI Act (which was repealed and replaced shortly afterwards) highlight the growing tension between national AI policy and state-level governance.

California AI training data transparency

California’s Generative Artificial Intelligence Training Data Transparency Act requiring GenAI developers to publish high-level training data summaries is already facing legal challenge.

Dark patterns and AI

We assess how AI systems that purportedly steer, manipulate or obscure user choice may invite scrutiny under deceptive trade practices laws and Section 5 of the FTC Act.

GenAI and privilege waiver

Recent US federal and state decisions suggest privilege may not be automatically waived when materials are disclosed into GenAI tools—but organizations still need safeguards. Under English law, we have analyzed similar GenAI privilege issues.

AI bias testing and privilege

A recent decision finding bias-testing materials were privileged and protected from disclosure, where attorneys were sufficiently involved in the testing process, raises important questions about how companies structure AI compliance reviews.

AI-driven cyber threats in the insurance industry

The New York Department of Financial Services (NYDFS) issued industry guidance to licensees regarding security measures they should consider taking “in a heightened cybersecurity threat environment” in light of the release of increasingly powerful AI models.

card-grid-item, aem-sourced
Inside Tech Law Blog
Insights, analysis and trends
https://nortonrosefulbright.us3.list-manage.com/subscribe?u=43b7c58a9a1fa3ef1ce018981&id=1dcc778f3f
card-grid-item, aem-sourced
Data Protection Report
Data protection legal insight at the speed of technology
https://www.dataprotectionreport.com/subscribe/

Contact