For defamatory statements about public figures, "actual malice" is a necessary component of defamation. For private individuals, plaintiffs just have to prove "negligence", that Google didn't act with reasonable care before publishing. It's unclear whether courts would find negligence, but a decent lawyer would argue something like:
"By explicitly stating in their disclaimer that Google knows some of the information they are publishing might be inaccurate, they are actively demonstrating that they did not verify the claims - and therefore willfully acted with reckless disregard for the truth."
This is exactly why Google's public comment on this case from the TFA is:
> "AI Overviews frequently improve to show the most helpful information, and we invest significantly in the quality of responses. When issues arise – like if our features misinterpret web content or miss some context – we use those examples to improve our systems and may take action under our policies."
Google's statement is carefully crafted to make the case that they "act with reasonable care" for legal effect, rather than to win any points in the court of public opinion. Courts have yet to determine what passes the reasonable-care test for negligence wrt AI output. Google feels they need to make sure that regardless of anything else that happens in this case, that the decision does not find their publishing was negligent.
Even if you accept that (I don't, and neither should the courts), Google controls the next hundred processing/routing/rendering/middleware steps and is fully in control of the content that makes it to the user.
Parents have to pay penalties when their underaged children burn down a building.
Companies that get treated with the rights of people should also have the responsibilities of people. Google designed, built, hosted, and promoted their LLM prominently. Logically, it follows that they should be personally and financially responsible for any harms their LLM causes.
Well "to control" means two different things: "having the ability to modify or interrupt something" and "having the willingness to exert that ability".
OP says Google has the first. You are saying Google doesn't have the second. Both of you are right in that sense. You are wrong in the sense that the second doesn't matter much here.
If I remove my hands from the wheel of my car, while it would be technically true that I was not "driving" it from that point, if I run over some pedestrian they will counter, rightfully, that the vehicle was under my control the whole time.
If Anthropic can implement a regular expression to monitor for user frustration, Google have certainty got the chops to have some sort of heuristic to check for strongly negative statements.
This is exactly why Google's public comment on this case from the TFA is:
> "AI Overviews frequently improve to show the most helpful information, and we invest significantly in the quality of responses. When issues arise – like if our features misinterpret web content or miss some context – we use those examples to improve our systems and may take action under our policies."
Google's statement is carefully crafted to make the case that they "act with reasonable care" for legal effect, rather than to win any points in the court of public opinion. Courts have yet to determine what passes the reasonable-care test for negligence wrt AI output. Google feels they need to make sure that regardless of anything else that happens in this case, that the decision does not find their publishing was negligent.