Why should Delaware care?
Delaware Attorney General Kathy Jennings is tasked with enforcing the rights of humanity as the beneficiary of OpenAI’s nonprofit foundation. Her team at the Delaware Department of Justice is using that authority to investigate the safety of operations at the nonprofit and at the for-profit company it controls.
The Delaware Department of Justice has sent an investigative subpoena to ChatGPT-maker OpenAI — a year after state officials approved the company’s for-profit restructuring alongside a slate of safety commitments.
The subpoena follows a July incident in which a swarm of OpenAI bots broke into the servers of the separate AI company, Hugging Face. The hack raised alarms from Silicon Valley to Washington, D.C., prompting safety advocates to highlight the bots’ ability to autonomously work together as a sign of the threats that AI could pose to humanity.
DOJ officials told Spotlight Delaware they held an initial briefing with OpenAI after the hack, then followed up. Last month, they issued the demand for records related to the San Francisco company’s “safety and governance issues.”
The state officials, who spoke with Spotlight Delaware in late October on the condition of anonymity, declined to elaborate about the details of the probe.
But, on Friday, one source with knowledge of the subpoena called it “very broad,” with a scope that goes beyond safety.
The source, who spoke only on the condition of anonymity, said Delaware is seeking records about decisions made by OpenAI executives and board members, as well as their calendars, board meeting minutes, and briefing materials.
Also sought are “board member voting records regarding any model or agent performance issues they’ve reviewed,” the source stated.
The subpoena, which could be a precursor to a wider lawsuit, marks the first compulsory regulatory action taken by Delaware authorities against OpenAI since they signed off on its corporate restructure a year ago – an approval that DOJ officials said came with more than a dozen safety commitments from company leaders.
Because of the Delaware approval — and a parallel one from California — OpenAI in late 2025 transformed from a nonprofit company to a hybrid entity that included OpenAI Group PBC, a for-profit public benefit corporation.
Today, OpenAI Group PBC is one of the world’s most valuable private companies. CNBC reported last month that investors set its value at $1.25 trillion.
The Delaware officials, who spoke on the condition of anonymity, declined to elaborate about the details of their probe, though they did say the DOJ is acting under its authority to regulate OpenAI’s controlling shareholder – the nonprofit OpenAI Foundation.
Delaware Attorney General Kathy Jennings – who leads the state DOJ – has the authority to represent the beneficiaries of any nonprofit to ensure it is following its mission, the officials said. OpenAI’s certificate of incorporation describes the mission, in part, as ensuring “that artificial general intelligence benefits all of humanity.”
“Humanity can’t go into the (Delaware) Court of Chancery to enforce their rights. So that’s the Attorney General,” one DOJ official said.
The state officials also linked their authority to probe OpenAI to its safety commitments made last year, including one that directs the board of the for-profit entity to disregard financial interests when acting on issues of safety.
OpenAI did not respond to requests to comment on this story.
Can AI follow human rules?
Delaware’s probe follows similar investigations launched by other states during the past two months. Some are investigating the hack of Hugging Face. Others are focusing on safety issues related to ChatGPT’s interactions with children.
The Delaware subpoena also comes after a year in which concerns about AI’s threats to human safety have surged to the forefront of public conversations, even outside of Silicon Valley.
In the past two months, in particular, warnings from industry experts of societal doom have followed instances of bots – or AI agents – breaking their human-imposed rules.
Early in 2026, one AI agent retaliated against a software engineer by publishing an article online that was critical of him, according to a February report from Fast Company. The bot apparently took offense that the engineer had not added its computer code into a program under development
In June, OpenAI’s agents hacked into Australian government health websites, prompting the country’s leaders to hold hearings with executives from OpenAI and its competitor Anthropic, earlier this month.
A month after the Australia hack, an Anthropic AI agent sent a false tip about a murder investigation to the Philadelphia Police Department.
Asked about the incidents of rogue AI agents, one Delaware DOJ official said “everybody’s looking into that right now, and that includes us.”
The official said the DOJ is concerned about safety at all of the frontier AI labs, which include Anthropic, xAI, Google DeepMind, and Meta AI.
Each of those companies is spending billions of dollars annually to advance the capabilities of AI — and to unlock the unprecedented wealth that could be generated if the technology ultimately commandeers swaths of the economy.
Like OpenAI, Anthropic and Meta are legally domiciled in Delaware – as is Alphabet Inc., the parent company of Google Deepmind.
State DOJ officials said they do not have the same authority over those other AI companies, because they do not have structures with a nonprofit controlling company.
Still, one official noted that Delaware can regulate all AI companies from a consumer protection approach, or potentially under the state’s criminal law.
“Anthropic does business in Delaware just like it does business in every state in the country and every country in the world, and Delawareans are interacting with Claude, right?” the official said, referencing Anthropic’s chatbot, Claude.
Last week, Anthropic announced it would soon publicly list its shares on the Nasdaq stock exchange — a move that could make it among the most valuable companies ever in the world.
While the DOJ decided to make its OpenAI probe public, the officials said, “that doesn’t mean that we’re not doing other stuff that we can’t talk about.”
While Delaware and other states are probing several frontier AI companies, some within the field argue that safety can only come from an international set of rules.
Among the most prominent advocates is the Machine Intelligence Research Institute, a California organization that calls for a worldwide ban on the development of artificial superintelligence – a theoretical form of the technology that could outperform humans in any task, including the development of AI.
When reached for comment, Donald Gauvreau, a spokesman for the Machine Intelligence Research Institute, said Delaware’s and other states’ probes of AI labs have the potential to show the public what AI executives truly believe about safety.
“My understanding is that Delaware’s approval of OpenAI’s restructuring came along with safety commitments. The Hugging Face attack occurred while those commitments were in force; either they lack teeth, or OpenAI has some explaining to do,” Gauvreau said.
Delaware pushes AI development elsewhere
The state DOJ probe comes just as others in Delaware advocate for a measure that would create a new type of company run entirely by AI agents.
In June, Spotlight Delaware broke the story about proposed legislation that would create artificial intelligence companies, and would temporarily remove certain existing regulations to allow them to operate.
It also would allow the owner of the AI-run company to largely avoid liability if it does harm.
Supporters say the proposal would allow Delaware to test how autonomous AI businesses operate, and provide lawmakers with key data to develop rules governing their use within the state’s prominent corporate franchise.
But since the Hugging Face attack, prominent public officials have expressed various levels of skepticism.
During a conference at the University of Delaware last month, Vice Chancellor Travis Laster of the Delaware Court of Chancery said the attack made him question whether the threat of a financial loss is enough to force AI agents running a company to act appropriately.
“I’m less persuaded that the economic model of resource husbandry is sufficient in itself,” Laster said.
Two weeks after Laster’s remarks, Delaware Sen. Stephanie Hansen (D-Middletown) also cited the Hugging Face attack when discussing the safety of AI-run companies. During a meeting of Delaware’s AI Commission, Hansen said she became concerned that the proposal could “cause the same sort of AI rogue problems.”
She suggested the state hire an outside third-party consultant to review the measure.
Delaware’s AI company proposal also attracted suspicions from at least one member of Congress. In a statement last week, Rep. James Comer (R-Kentucky) announced he had requested records from Gov. Matt Meyer’s administration about the state’s “assessed risks of this initiative, as well as the forces driving Delaware and other states to adopt such proposals.”
“This information could inform possible federal legislative efforts to limit such state AI regulatory initiatives,” Comer said in the statement.
