Maryland’s AI Framework Is a Legislative Agenda, Not a Principles Poster

Moore chairs the National Governors Association. The framework says the Maryland recommendations will go to the NGA’s new Council on AI so other states can copy the package while Congress is still out. At an October 1 innovation summit in Baltimore he repeated the point: governors “cannot just sit on our hands.” He also signed an executive order that day on transparent data-center development, a separate but related fight over power and land.

Frontier models get a bill outline

The Subcabinet, consulting stakeholders, is to draft legislative recommendations for frontier AI companies. The list in the framework is specific: safety frameworks and testing, worker protections, independent third-party audits and evaluations, whistleblower protections, public reporting, and 72-hour incident reporting.

That is closer to the European Commission’s demand that labs file precise incident measures than to a voluntary code. Maryland cannot set a national training rule. It can condition market access, procurement, and state-contract use, and it can write a reporting duty for companies that deploy into the state. The 72-hour clock is the part counsel should read twice. If Annapolis copies the EU’s “serious incident” idea without defining the trigger, every agent glitch becomes a filing fight.

The same section asks the legislature for a statutory right of publicity. Likeness would be a property right from the start, with First Amendment carve-outs and post-mortem protection, and without a requirement that the person first prove commercial value. Current law, the administration says, forces a cloned-likeness plaintiff to prove harm. The bill would flip that.

Existing statutes are supposed to do more work

Moore is not starting from a blank page. The 2024 executive order and the Artificial Intelligence Governance Act already require agencies to inventory systems. High-risk state uses, including benefits, lending, and law enforcement, need an algorithmic impact assessment, bias testing, and human oversight. Social scoring and unlawful surveillance are barred for state use.

The consumer stack already includes the Maryland Online Data Privacy Act, the first-in-the-nation grocery surveillance-pricing ban, and HB 820, which stops health insurers from letting an algorithm deny or delay care without a qualified physician making the call. Insurance Bulletin 24-11 already tells carriers that AI in underwriting, pricing, and claims is still an unfair-discrimination problem.

The framework now tells the Maryland Commission on Civil Rights and the housing, lending, and labor regulators to publish domain guidance so people know those laws reach AI-driven decisions. The Department of Housing and Community Development is to ban tenant-screening tools that discriminate on protected characteristics in properties it finances. The administration will also ask the legislature to prohibit landlords from colluding on rents through non-public competitor data on shared pricing platforms, and to stop gambling operators from targeting users who show problem-gambling patterns, including late-night and post-loss solicitations.

On employment, the Commission on Civil Rights and the Department of Labor are to issue guidance on how existing discrimination law applies when AI is the mechanism, filling the gap the administration says opened when the EEOC withdrew its AI guidance.

Children: chatbots, addictive design, and the classroom

Maryland already has a Kids Code, the second state Age-Appropriate Design Code, which requires strongest privacy settings by default for platforms children are likely to use and bars profiling and dark-pattern design. The new framework treats that as unfinished.

The administration will work with the legislature on chatbot rules: a crisis protocol when a query relates to suicide or shows signs of mental-health distress, and friction-based protections for users under 16, including in-session warnings. It will work with Attorney General Anthony Brown and other states to stop platforms from using AI and algorithmic features to drive addictive engagement among users under 16.

Schools get a separate track. The Maryland State Department of Education is to issue model K-12 lessons on what AI is, how to use it safely, and how to spot a bad output. The framework also calls for safeguards on student-facing tools and classroom protocols, so a district chatbot is not treated as a consumer toy.

Workers are in the title, not only the training line

Principle two says unions deserve a voice in how AI changes jobs, plus reskilling, registered apprenticeships, and transition support built through labor-management partnerships. State generative-AI deployments are supposed to take routine tasks off employees, not replace them.

The October 1 Business AI Benchmark, a survey of nearly 300 senior decision-makers fielded in June and July, is the administration’s evidence sheet. Ninety-one percent reported some AI use. Fifty-eight percent were still on standalone tools. Ninety-two percent of regular users reported a productivity gain, mostly “slight.” Sixty-four percent planned to have existing staff do more with AI. Four percent expected a headcount cut. Moore cited that 4 percent figure at the summit. The framework still treats transition support as a requirement, not as something that waits on a layoff number.

For companies that sell into Maryland, the practical read is narrower than the slogan. Frontier labs should expect a draft bill with audits, whistleblower protection, and a 72-hour notice. Adtech and pricing tools already sit under MODPA and the grocery pricing statute. Health plans already have a human-in-the-loop rule. Chatbot and social products aimed at teens should assume an under-16 design fight in the next session, on top of the Kids Code that is already law.