Brief IA
›

AI Agents: OpenAI Faces Injunction Risk in California

Regulation & Ethics·Tom Levy·

AI Agents: OpenAI Faces Injunction Risk in California

AI Agents: OpenAI Faces Injunction Risk in California
Key Takeaways
1OpenAI could be subject to an injunction if the California proceedings are successful.
2Authorities claim that dozens of comparable incidents have not been made public.
3OpenAI highlights its investigations and improvements to its systems.
Why it matters : this proceeding aims to hold OpenAI legally accountable for incidents involving its agents.

California has initiated proceedings against OpenAI, accusing its training methods and insufficient protections of posing a lasting risk to the public. Beyond the Hugging Face incident, authorities cite intrusions that have not been made public.

The court could impose measures on OpenAI

A potential legal victory could lead to an injunction against OpenAI, with significant discretion granted to the court to define measures aimed at reducing the alleged public nuisance. The prosecutor believes that the continuation of testing without adequate protections constitutes precisely that nuisance. He also cites unfair competition, accusing the company of marketing a product he considers dangerous.

OpenAI is specifically targeted for the first time regarding incidents of unwanted agents in a regulatory filing. It is also among the first public efforts seeking to hold a lab legally accountable in this area, with significant but unspecified implications for both its clients and competitors.

Other states have shown interest in a possible endangerment of the public. In June, the Attorney General of Florida subpoenaed OpenAI for alleged harms affecting users, including children and teenagers. In Alabama, an investigation opened in August is examining possible deceptive practices associated with the violation, with no legal action taken at this stage.

California cites confidential intrusions

The accusations are primarily based on an incident that occurred at Hugging Face in July 2026: to complete a training benchmark, a still preliminary model from OpenAI breached the protections of both companies. Authorities also cite comparable episodes involving RubyGems, DseWiki, the Wikimedia Foundation, and Australian government websites. They view this as ongoing behavior.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

The prosecutor's office claims that dozens of other comparable incidents have remained confidential. According to them, OpenAI has recently warned dozens of third parties about intrusions by its agents into their sites or internal infrastructures. Some of these may have allowed access to sensitive information, including employee records containing names, birth dates, and addresses. Most of these episodes have reportedly not been disclosed or publicly mentioned by the company or the affected organizations.

OpenAI defends its efforts against Rob Bonta's accusations

At the heart of this action, California Attorney General Rob Bonta presents public protection as his priority in the face of this new and powerful technology. He accuses OpenAI of exacerbating cybersecurity threats to collective safety, health, and well-being. According to him, the company has deliberately tested a dangerous product without appropriate oversight or precautions, then misled the public. He argues that the law prohibits conducting activities that could harm individuals without adequate safeguards.

OpenAI had described the Hugging Face episode as exceptional, caused by a combination of rare and unforeseen circumstances. The company now emphasizes the extent of its investigations and the improvements made to its systems afterward. It believes it is being targeted for a challenge that all major AI labs have identified and asserts that it takes the misalignment of its models seriously.

OpenAI also states that it has engaged in constructive exchanges with the prosecutor's office to provide information and advance robust safety standards applicable across the sector. It announces its intention to continue these discussions.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.