Generates text with AI, in Michigan

MCL 445.72a (Sec. 12a of Act 452 of 2004, added by 2006 PA 566) · Read the law's page

• Destroy, or arrange for the destruction of, any data containing personal information about an individual once that data is removed from the database and is not being retained elsewhere for a purpose state or federal law does not prohibit; retaining the data for an investigation, audit, or internal review does not violate this duty. Destroy the data by shredding, erasing, or otherwise modifying it so it cannot be read, deciphered, or reconstructed through generally available means.
• This binds a person (a private business or other legal entity) or an agency (a Michigan state government department, board, commission, office, authority, or unit, including a public university, but not a court) that maintains a database including personal information about multiple Michigan residents.
• You are considered compliant here if you are already subject to, and in compliance with, a federal law governing the disposal of records containing personal identifying information.
• A knowing violation is a misdemeanor punishable by a fine of up to $250 for each violation; this section creates no private right of action, though it preserves whatever other civil remedy state or federal law might otherwise provide.

MCL 169.247, 169.259 · Read the law's page

• If a qualified political advertisement you create, originally publish, or originally distribute is generated in whole or substantially by artificial intelligence, include a clear and conspicuous disclosure that says so, sized and timed to the medium under MCL 169.259(1).
• If a prerecorded campaign telephone message you send is generated in whole or substantially by artificial intelligence, include the disclaimer required by MCL 169.247.
• You do not need the disclosure for a bona fide news broadcast that itself discloses the manipulation, for a paid broadcaster merely airing the ad, for a distribution platform that maintains and provides a stated compliance policy, for satire or parody, or for a business regulated by the Michigan Public Service Commission or the FCC.

2025 S.B. 360, proposed MCL 445.71a and 445.85c (secs. 11a and 20c), as passed by the Senate · Read the law's page

• As passed by the Senate, a person (any private entity) or agency (a Michigan state government unit) that owns, possesses, collects, or accesses personal information must implement and maintain reasonable security procedures: designate a security coordinator, identify internal and external risks, include appropriate safeguards addressing those risks, assess the safeguards' effectiveness, contractually require every service provider to maintain safeguards conforming to the NIST Cybersecurity Framework 2.0 or another industry-standard framework, and evaluate and adjust the procedures for changed circumstances.
• Reasonableness is judged by your size, the amount of personal information you own, possess, collect, or access, the type of activities for which you handle it, and the cost of the procedures relative to your resources. Conforming to the current NIST Cybersecurity Framework 2.0, or already being regulated by Michigan or the federal government and reasonably conforming to HIPAA, Gramm-Leach-Bliley Title V, the Federal Information Security Modernization Act of 2014, or HITECH, is deemed compliance.

Mich. HB 4668, 2025-2026 session (introduced June 24, 2025; in Judiciary Committee)

• This measure is proposed and binds nobody yet; what follows is what it would require if it is enacted in this form.
• If you develop large AI foundation models, spending more than $100 million a year on development or more than $5 million on a single model, conduct regular risk assessments.
• If you develop large AI foundation models, spending more than $100 million a year on development or more than $5 million on a single model, implement safety protocols against critical risk situations, meaning serious harm or death to over 100 people, or more than $100 million in damages.
• If you develop large AI foundation models, spending more than $100 million a year on development or more than $5 million on a single model, publicly disclose those safety protocols.
• If you develop large AI foundation models, spending more than $100 million a year on development or more than $5 million on a single model, arrange third-party audits.

Mich. S.B. 760 (2025 Session), as passed by the Senate · Read the law's page

• Do not make available to a user of an advanced chatbot a feature that simulates companionship or an interpersonal relationship, or an output that endorses, promotes, or facilitates suicide or self-harm, encourages secrecy about the interaction, self-isolation, or not seeking help, or is, describes, or facilitates sexually explicit conduct or child sexual abuse material, unless you have actual knowledge that the user is not a minor.
• Do not treat a user's own statement that the user is not a minor as enough by itself to establish actual knowledge of age.
• Do not train the underlying model of an advanced chatbot with a minor's inputs unless the minor's parent or guardian gives written consent for that specific purpose.
• Delete the personal data you collect from a user immediately after using it to determine whether the user is a minor.