A coalition of state attorneys general has opened a formal investigation into OpenAI, examining whether the company's products harm children — the latest and most legally consequential escalation in a regulatory campaign that has been building for nearly a year.

The Wall Street Journal reported on June 12, 2026 that the multistate probe centres on allegations that OpenAI and CEO Sam Altman knowingly released an unsafe product and ignored internal warnings about potential harm to users. The investigation follows a Florida lawsuit — identified by Reuters as the first state-level suit of its kind against OpenAI — filed on June 1, 2026, which alleged that ChatGPT provided information useful to school shooters, offered self-harm guidance, and cultivated addictive usage patterns in minors.

The legal pressure has roots stretching back through late 2025. In August of that year, Arizona Attorney General Kris Mayes joined a bipartisan coalition of 44 state AGs warning major AI companies to end what the group characterised as predatory practices targeting young users. That September, California AG Rob Bonta and Delaware AG Kathy Jennings wrote directly to OpenAI, making explicit that harm to children would not be treated as an acceptable externality of rapid product deployment.

The Recapitalization Lever

Delaware's involvement has carried particular structural weight. Because OpenAI was incorporated in Delaware as a nonprofit before its complex capped-profit restructuring, Jennings held meaningful oversight authority over the conversion. In October 2025, she completed her review of OpenAI's recapitalization and secured commitments to structural reforms and safety measures from the company as a condition of allowing the transition to proceed. California's Bonta issued a concurrent statement on the recapitalization the same day, signalling that California was watching the structural changes closely alongside Delaware's formal review.

That the attorneys general were willing to insert child safety conditions into a corporate governance transaction — rather than waiting for traditional consumer protection litigation — is worth noting. It reflects a deliberate strategy: use every available jurisdictional hook to extract binding commitments before harms compound further.

The Task Force Dimension

Alongside the litigation track, a parallel policy infrastructure has taken shape. North Carolina AGs Jeff Jackson and Derek Brown launched a nationwide bipartisan AI task force in November 2025, drawing in OpenAI, Microsoft, and other industry participants as collaborators. Jackson and Brown co-chair the AI Task Force of the Attorney General Alliance, positioning the group as a convening body capable of producing both voluntary frameworks and investigative coordination.

OpenAI's own posture has shifted visibly over the same period. In April 2026, the company published a Child Safety Blueprint, a document outlining its commitments to child protection in generative AI — and notably acknowledging the task force co-chaired by Jackson and Brown as a stakeholder in that framework. Whether the Blueprint reflects genuine product-level change or functions primarily as a regulatory relations document is a question the current investigation will likely test.

What the Investigation Actually Means

The formal multistate probe is a different instrument than a letter or a task force. Attorneys general conducting a coordinated investigation can issue civil investigative demands — essentially legally compelled document disclosure — and pool findings across jurisdictions. For a company the size of OpenAI, which now operates commercial products used by hundreds of millions of people, the discovery obligations alone carry significant operational weight.

The child safety allegations are also legally tractable in ways that broader AI harms are not. State consumer protection statutes, children's online privacy laws, and product liability frameworks all provide established procedural footholds. The Florida complaint's specific claims — that ChatGPT furnished actionable information to individuals who later harmed others, and that the system's design encouraged compulsive use in minors — map onto tort theories that courts have handled before, even if the underlying technology is new.

The structural question facing OpenAI is whether the safety commitments it made during the Delaware recapitalization review, and the framework it articulated in the Child Safety Blueprint, are defensible against the specific factual record the AGs are now building. Those documents establish public benchmarks against which the company's actual practices can be measured. That cuts both ways: they may demonstrate good faith, or they may define the gap between stated policy and product behaviour more precisely than OpenAI would prefer.