Unlike Facebook, AI Companies Write the Words Themselves
Because chatbots write their own answers, insurers and regulators see a liability gap just as big AI labs head toward IPOs.
WSJ What's News · AI Has an Insurance ProblemThe legal shield that protected Facebook from lawsuits over what appeared on its site may not protect makers of artificial intelligence, and the reason comes down to who does the writing.
That earlier shield is Section 230, a legal protection that gave companies like Facebook, in the words of Wall Street Journal reporter Jeanne Eaglesham, "quite a lot of immunity over being sued over the content on their websites." The logic was straightforward. Users posted the words, photos, and videos, and the company just hosted the posts.
An AI chatbot works differently. When a user asks a question, the system generates the answer itself. That difference matters, said Eaglesham, citing Moody's Analytics, "because the companies themselves are contributing to the content." If the company helped create the words, it cannot as easily claim it was just hosting someone else's speech.
That leaves AI labs more exposed to being sued for what their systems produce. Recent episodes cover a wide range, including AI agents that hacked websites, chatbots allegedly encouraging people toward suicide, and AI hallucinations cited by law firms. The Journal's reporting put the potential damage from such emerging threats at amounts in the tens of billions of dollars, and possibly much higher.
The exposure is getting more attention because large AI labs are approaching public stock offerings. Those offerings are drawing closer examination of the financial safety nets at companies like OpenAI and Anthropic, which hold vast cash reserves and could pay for some of those claims out of their own funds. But regulators, lawmakers, and investors are also looking at insurance as reassurance, as a sign that an outside company has examined the risks and is willing to underwrite them.
Questions people ask
What does Section 230 actually do?
Section 230 is the part of U.S. communications law that has generally kept online platforms from being treated as the publisher of material posted by users. In practice, it has let social media companies argue that the person who posted a defamatory statement or other harmful content, not the platform, is legally responsible.
Have courts decided whether Section 230 covers AI outputs?
Courts are still working through that question, and there has been no definitive Supreme Court ruling that AI generated text is either covered or excluded. Lower courts have reached different conclusions in early cases involving autocomplete, recommendations, and generative tools, so lawyers treat the protection for AI as uncertain.
Why can't AI labs just buy regular business insurance?
Standard business policies for general liability and cyberattacks were written before generative AI existed, so it is often unclear whether AI harms count. Insurers are also considering excluding some AI risks, and startup insurers that specialize in AI remain small relative to the scale of potential claims.