Former US regulator says AI companies not exempt from current law

Official patience with artificial intelligence companies allowing their models to disrupt and interfere with other organisations’ systems is wearing thin, with former chair of the United States Federal Trade Commission, Lina Khan, saying in public that they have no exemption from existing law.

“Law enforcers already have authority to charge companies and their CEOs for creating and releasing dangerous, unvetted, or defective products,” Khan wrote in a post on the X social network.

She argued that discussions over new regulatory regimes for AI should not distract from enforcing the rules already in force, adding “… there’s no AI exemption from laws already on the books.”

Khan chaired the FTC from 2021 until January 2025, with her tenure marked by an aggressive antitrust and consumer protection stance towards ‘big tech’.

Under the US FTC Act and equivalent American state laws, shipping flawed AI tools without adequate safeguards to detect or stop defective agents could amount to an “unfair or deceptive” act.

She added that some US state attorney-generals were already exploring criminal liability for AI companies and their chief executives, but did not name the states in question.

US competition law could also apply, and Khan pointed to the concentrated and interconnected ownership of the sector ties that can blunt accountability.

Investment in the AI sector, particularly in large vendors such as OpenAI and Anthropic, is dominated by tech giants Nvidia, Microsoft, Amazon and Google.

Khan referred to the highly publicised incident in July this year which saw OpenAI models attack machine learning forum Hugging Face, saying the company could face liability.

However, Nvidia agreed to buy Hugging Face earlier this month, which means any legal action is unlikely to take place, given the AI and graphics card vendor’s “strong incentive to see OpenAI continue full speed ahead,” Khan added.

AI agents incidents continue

OpenAI has added to the growing list of cases where its models may have bypassed third parties’ security controls, impaired availability of online services or negatively impacted them

The company said it has notified “dozens of third parties” about cases involving the above behaviour.

Although some names and identifying details are omitted for protection, OpenAI owned up to its agents using the RubyGems software package manager platform to access the Internet.

The agents did this to carry out what OpenAI said are benign tasks and to retrieve public information.

OpenAI said it has not been able to verify the agents uploaded malicious packages in the incident which took place in May this year.

That denial runs counter to a report by three researchers – Spencer Kitts, Thomas Larsen and Sydney von Arx – which stated that not only did the OpenAI agents write hundreds of malicious packages, but also the AI uploaded them to RubyGems as well.

Furthermore, the agents also tried to steal user RubyGems API keys in the attack, which was dubbed the “GemStuffer campaign” by security vendor Socket.

The agents also abused the RubyDoc.info site, which builds and hosts documentation for RubyGems packages, to run arbitrary code.

It’s not clear at this stage what the OpenAI agents tried to achieve with the actions they took, which appears to be attacking RubyGems to access publicly available data.

We’re dealing with a major malicious attack on @rubygems right now. Signups are paused for the time being.

Hundreds of packages involved – mostly targeting us, but some carrying exploits. The team has been on this for hours. More details to follow once we’re through it.#ruby

— Maciej Mensfeld (@maciejmensfeld) May 12, 2026

As a consequence of the attacks, RubyGems had to investigate the incident and halt sign-ups temporarily.

OpenAI also admitted that its agents had written to a wiki, posting a large volume of messages (over 18,000) so as to collude with each other for web lookup tasks.