280 State AI Laws in One Year: Why Federal Deregulation Won't Lighten the Load
States passed 280 AI laws this year while Washington deregulates. Colorado's SB 26-189 and the EU AI Act bring new duties for AI used in hiring from January 2027.

Nathan Brooks
· 4 min read
American lawmakers passed 280 AI laws this year alone, and that count came with weeks still left on the legislative calendar, according to the National Conference of State Legislatures. For HR leaders, the message is blunt. Federal deregulation does not translate into fewer compliance duties for employers deploying AI in hiring, evaluation and other consequential decisions.
The picture in Washington suggests relief. The administration has made reducing regulatory barriers a priority, and in December 2025 the White House issued an executive order seeking to limit conflicting state AI regulations. But an executive order does not automatically invalidate state laws. States keep legislating, and several new statutes reach directly into employment processes.
Which new rules touch HR technology?
Colorado enacted SB 26-189 in May 2026, replacing its earlier AI law with a framework governing automated decision-making technologies used in consequential decisions, including employment, lending and housing. Beginning January 2027, covered businesses face disclosure and other obligations, while consumers gain rights concerning inaccurate personal data.
Other measures add to the stack:
- California's Transparency in Frontier Artificial Intelligence Act (SB 53), effective this January, requires covered frontier AI developers to publish safety frameworks and report critical safety incidents.
- The New York RAISE Act establishes similar obligations beginning January 2027.
- California Governor Gavin Newsom signed 11 AI safety bills in September alone, capping a legislative year that saw nearly 30 AI-related measures enacted in the state.
- Illinois enacted the Artificial Intelligence Safety Measures Act (SB 315) in July 2026, effective January 2027, requiring safety frameworks, incident reporting and, for the largest developers, independent compliance audits.
- Texas TRAIGA, in effect this year, establishes an intent-based liability model for AI development and deployment in the state.
- Colorado's HB 26-1263 adds disclosure requirements and protections for minors for conversational AI services.
Multinational employers face an additional layer. The EU AI Act entered active enforcement, and in July 2026 the EU adopted the AI Omnibus amendments, simplifying certain requirements and pushing deadlines for high-risk systems into 2027 and 2028. The core framework holds, with obligations on transparency, prohibited practices, general-purpose models and high-risk applications.
Why can't employers certify once and move on?
Consider a company using AI to screen job applicants, evaluate customer eligibility for financial products or automate service interactions. The same application can face different requirements depending on where it operates, which individuals it affects and how its outputs influence decisions.
Even inside one organization, an identical model might power both a low-risk productivity assistant and a highly consequential employment tool. The model is the same. The risks and regulatory obligations are not. That is why a single certification cannot cover every downstream use, and why HR teams must understand how each AI system is used, what data it touches and which decisions it shapes.
Regulatory fragmentation also exposes a visibility gap. Enterprises may approve a handful of official AI platforms while individual departments independently integrate models into customer service, analytics, software development and decision workflows. Vendors keep embedding AI into existing products, often expanding the footprint without corresponding governance. Shadow AI compounds the problem as workarounds multiply.
Agentic AI raises the stakes further. These systems can access sensitive information, interact with enterprise applications and take actions rather than merely generate responses. When something goes wrong, organizations must be able to determine which system was involved, what data it accessed, how it behaved and which safeguards were in place.
What does a workable compliance foundation look like?
Building separate compliance processes for every jurisdiction is unsustainable given the pace of legislation, the NCSL's 280-law figure suggests. A common enterprise-wide foundation for responsible AI, with added safeguards where specific rules or use cases demand them, is the more durable path.
That foundation starts with an accurate inventory of AI applications, models, agents and third-party integrations. Pre-deployment testing remains necessary but insufficient. Changes in prompts, data, integrations or underlying models can introduce new risks after launch, so organizations need continuous monitoring for accuracy, reliability, security, privacy, bias and adherence to business policy, with documented evidence of testing and remediation.
Governance must also extend beyond the model itself. A reliable system can still create serious problems when connected to sensitive databases, external tools or automated workflows without appropriate access controls. Policy documents alone do not equal responsible AI; principles around fairness, transparency and accountability must translate into measurable technical controls and operational processes.
Federal lawmakers may eventually pass a national framework that supersedes some state requirements, and courts may reshape the terrain. Even so, multinationals will still face distinct international obligations, and employers that know where their AI operates, continuously evaluate its behavior and can demonstrate oversight will be best positioned as the rules keep multiplying.
Source: cio.com
More from Nathan Brooks
Show full bio
News editor covering people, launches and day-to-day practice in Recruiting and talent acquisition, learning and development at HR Briefing.
270 articles