News-May-22-2026

May 22, 2026 — State AI legislative activity this week centered on youth-safety, chatbot, neurological-rights, and crossover-deadline bills; President Donald Trump postponed an AI executive-order signing ceremony after saying he disliked aspects of the draft order; POLITICO agreed to shut down two AI tools after a landmark labor arbitration involving the POLITICO and E&E News Guild; the Department of Health and Human Services announced expanded AI screening of federal health-program audits; federal prosecutors brought early Take It Down Act charges over AI-generated sexual deepfakes; California Governor Gavin Newsom signed an executive order directing state agencies to prepare for AI-related workforce disruption; the National Transportation Safety Board suspended public access to its accident-docket system after AI-assisted reconstructions of cockpit audio raised federal privacy-law concerns; and Estonia authorized Bliq.ai to begin fully driverless road operations under remote supervision in what Tech.eu described as the first authorization of its kind in an EU member state.

Contents

1. State AI legislative roundup
2. Trump postpones AI executive order signing
3. POLITICO agrees to shut down AI tools after arbitration
4. HHS expands AI audit-fraud screening
5. Federal prosecutors bring early Take It Down Act deepfake charges
6. California orders AI workforce-disruption planning
7. NTSB restricts accident-docket access after AI cockpit-audio reconstructions
8. Estonia authorizes Bliq.ai fully driverless road operations

---

State AI legislative roundup

Minnesota legislators gave final approval to HF 4138 before their May 17 adjournment, and the youth social-media safety measure was sent to Governor Tim Walz on May 20, according to the Transparency Coalition's weekly legislative update. South Carolina Governor Henry McMaster signed H 4591, the Stop Harm From Addictive Social Media Act, on May 19, and the act requires covered platforms to verify account-holder ages, obtain parental consent for minors, and apply default minor-account settings. Missouri lawmakers passed SB 1019 before adjourning sine die on May 15, and the health-care bill includes a prohibition on offering AI therapy chatbots while awaiting the governor's signature. Vermont Governor Phil Scott signed H 814, a bill recognizing personal neurological rights, while the legislature continued working beyond its original mid-May adjournment target.

California's annual appropriations "suspense" process left roughly 30 AI-related bills moving forward, with several Assembly bills crossing into Senate committees and several Senate bills crossing into Assembly committees. The California bills still moving include AB 1159 on student privacy protections for school-purpose digital operators, AB 1609 on customer-service chatbots, AB 1979 on AI in health-care services, SB 11 on social-media warnings and addictive feeds, SB 1119 on companion chatbots, and SB 1142 on the Digital Dignity Act. Arizona lawmakers recessed until June 1 amid a budget deadlock, leaving AI-related bills such as HB 2133 on synthetic intimate imagery, SB 1786 on provenance data for generative-AI media, and HB 2592 on state-agency AI adoption still in play rather than finally enacted.

The update shows a continuing state-level pattern in which child-safety, chatbot-governance, AI-disclosure, and worker or health-care uses of AI are advancing through ordinary state legislative procedures rather than through a single federal framework.

---

Trump postpones AI executive order signing

President Donald Trump said on May 21 that he postponed a planned signing ceremony for an artificial-intelligence executive order because he "didn't like certain aspects" of it, and CNBC reported that he said the United States was ahead of China on AI and that he did not want an order to impede that lead. CyberScoop reported that a circulating draft would have established a voluntary federal testing regime for frontier AI models, including a 90-day pre-release study period for covered models and classified evaluations by the National Security Agency. CyberScoop also reported that the draft contemplated roles for the Department of the Treasury, the Office of the National Cyber Director, the Cybersecurity and Infrastructure Security Agency, and the National Institute of Standards and Technology in information-sharing and model-vetting work.

The postponement is significant for AI law because it leaves the administration's proposed federal AI-security review process unresolved while state legislatures continue advancing AI-specific bills.

---

POLITICO agrees to shut down AI tools after arbitration

The Washington-Baltimore News Guild announced that POLITICO agreed to shut down the Capitol AI Report-Builder and not revive its Live Summaries AI feature after negotiations over a November 2025 arbitration ruling involving the POLITICO and E&E News Guild. The Guild said the arbitrator had found that POLITICO violated collective-bargaining-agreement provisions requiring notice, bargaining, and human oversight before deploying AI tools affecting journalists' work.

According to the Guild, Capitol AI Report-Builder produced branded policy reports for POLITICO Pro subscribers without editorial review, and Live Summaries generated unedited coverage of political events including the 2024 Democratic National Convention and vice-presidential debate. The remedy is significant for AI law because it treats contractual AI protections, bargaining duties, and journalistic accuracy obligations as enforceable limits on newsroom AI deployment rather than voluntary internal guidelines.

---

HHS expands AI audit-fraud screening

The Department of Health and Human Services announced on May 21 that it would use ChatGPT and other AI tools to analyze audit reports from all 50 states and other recipients of federal health dollars on an ongoing basis. AP reported that covered recipients include state Medicaid programs and federal grantees in research, addiction services, and other HHS-funded programs. AP also reported that recipients that fail to file required audits or resolve problems identified in them could face a loss of funding.

The initiative is significant for AI law because it applies generative-AI review tools to federal grant oversight, fraud screening, and potential funding consequences for state and institutional recipients of HHS money.

---

Federal prosecutors bring early Take It Down Act deepfake charges

Federal prosecutors charged Cornelius Shannon and Arturo Hernandez with using AI to create sexually explicit nude videos and photos under the Take It Down Act, according to AP. AP reported that Shannon, a New Jersey resident, allegedly published at least 240 albums of AI-generated pornography featuring female politicians, musicians, and singers, while Hernandez, a Texas resident, allegedly published deepfakes of celebrities and private women including recent high-school graduates. AP reported that the two defendants, who do not appear to be connected, are among the earliest defendants charged under the law and face up to two years in prison.

The charges are significant for AI law because they show federal prosecutors using the Take It Down Act's AI-deepfake provisions soon after enactment rather than treating the statute as only a notice-and-removal regime for platforms.

---

California orders AI workforce-disruption planning

California Governor Gavin Newsom signed a May 21, 2026 executive order directing state labor, economic-development, and technology agencies to assess how AI adoption may affect workers, businesses, hiring, layoffs, and skills needs. The order asks state officials to evaluate policy options for mitigating AI-related job displacement, including potential incentives for employers that adopt AI without replacing workers. StateScoop reported that California agencies must study labor-market shifts tied to AI adoption, including layoffs, hiring changes, and skills gaps.

---

NTSB restricts accident-docket access after AI cockpit-audio reconstructions

The National Transportation Safety Board temporarily disabled public access to its online docket system on May 21, 2026 while reviewing materials that allowed people to reconstruct approximations of cockpit voice recorder audio with image-recognition and other computational methods. Ars Technica reported that the reconstructions involved sound-spectrum imagery released in NTSB investigation materials, including the ongoing investigation of the November 4, 2025 UPS Flight 2976 crash in Louisville, Kentucky. The NTSB said it does not release cockpit audio recordings, and Ars Technica noted that a 1990 federal law prohibits the agency from publicly sharing cockpit voice or video recorder audio in order to protect air-crew privacy.

The development is significant for AI law because it shows a federal transparency system being curtailed after AI tools made it possible to infer protected audio from lawfully released technical records.

Estonia authorizes Bliq.ai fully driverless road operations

Estonia authorized driverless-car startup Bliq.ai to operate vehicles on public roads without a driver behind the wheel, allowing fully driverless road operations under remote supervision. Tech.eu reported that the authorization is the first of its kind in an EU member state and followed validation that included test-track sessions and real-world testing in Tallinn traffic with a safety driver onboard.

Bliq.ai said its deployment model upgrades existing software-defined vehicles with sensor and compute hardware rather than manufacturing purpose-built autonomous vehicles. The development is significant for AI law because it shows an EU member-state regulator permitting fully driverless operations under a remote-supervision model that Bliq.ai says it plans to expand to other European markets.