Latest / Bloomberg Talks / California AG Rob Bonta Talks OpenAI Lawsuit, AI Safety
Transcript
- 0:02Speaker 1: Bloomberg Audio Studios.
- 0:04Speaker 2: Podcasts. Radio.
- 0:06Speaker 3: News.
- 0:07Speaker 2: California's Attorney General is suing OpenAI for allegedly enabling at
- 0:12Speaker 2: least 80 rogue agent cyber attacks by unsafely developing its
- 0:16Speaker 2: AI products. The lawsuit accuses OpenAI of violating California's unfair
- 0:20Speaker 2: competition law by skirting safety procedures and making misleading public statements.
- 0:25Speaker 2: Attorney General of California, Rob Bonta, joins us now. for more.
- 0:30Speaker 2: And Attorney General, thank you for your time. I want
- 0:32Speaker 2: to start with the evidence. What evidence was presented to
- 0:36Speaker 2: you that OpenAI behaved, acted in a way that was,
- 0:40Speaker 2: to California's mind, illegal and prompted you to take.
- 0:43Speaker 1: Action, please?
- 0:46Speaker 3: Well, basically, there were multiple incidents, at least 80, where
- 0:51Speaker 3: open AI failed to take steps during the testing and
- 0:55Speaker 3: development of advanced AI models to safeguard and monitor its
- 1:00Speaker 3: own technology. And their technology, their AI agents, if you will,
- 1:05Speaker 3: escape their contained testing environment, got onto the Internet and
- 1:10Speaker 3: could go anywhere in the world. And they got onto
- 1:13Speaker 3: other systems and networks. The Hugging Face incident, obviously, very
- 1:16Speaker 3: well known, most infamous. They got onto websites as far
- 1:21Speaker 3: away as Australia. And they failed to take steps when
- 1:26Speaker 3: they had an advanced model that they knew was sophisticated, advanced, persistent.
- 1:31Speaker 1: And relentless.
- 1:32Speaker 3: And so they failed to keep Californians safe.
- 1:36Speaker 1: My job is to protect the health, safety, and welfare
- 1:37Speaker 1: of Californians. And that's why we brought the lawsuit.
- 1:41Speaker 2: You mentioned there the example of what happened in Australia.
- 1:43Speaker 2: And then, as you say, your jurisdiction, your office's remit
- 1:47Speaker 2: is to protect Californians. What is the clearest example of
- 1:51Speaker 2: harms to Californians from the activity that's been widely reported
- 1:56Speaker 2: and that OpenAI has also published its research on?
- 2:00Speaker 3: Well, the AD incidents show a variety of different examples
- 2:05Speaker 3: of OpenAI's own AI agents escaping their contained environment, getting into.
- 2:10Speaker 1: Other networks and systems.
- 2:12Speaker 3: And the harm is essentially unlimited. If they get into
- 2:18Speaker 3: a certain system or network, they could take down a
- 2:22Speaker 3: power grid, harm a water system, they could harm emergency response.
- 2:28Speaker 3: And so it's not our and coming on us to
- 2:30Speaker 3: wait until that happens and then take action. It's to
- 2:35Speaker 3: see all these signs that this is a risk that
- 2:38Speaker 3: is not something that is acceptable for Californians and take
- 2:41Speaker 3: appropriate steps.
- 2:43Speaker 1: It's a basic principle.
- 2:44Speaker 3: If you're going to test a dangerous product, then you
- 2:46Speaker 3: need to make sure you put proper safeguards in place.
- 2:48Speaker 3: It's been true for decades and for centuries, from aviation
- 2:52Speaker 3: to the railroads to chemical testing and beyond. It is
- 2:56Speaker 3: a simple principle. and open AI is failing to honor
- 3:02Speaker 3: that principle and is breaking the law in the process.
- 3:05Speaker 3: And for competition law, as you mentioned, also the public
- 3:07Speaker 3: nuisance law, visiting harm and great risk to California and Californians.
- 3:12Speaker 3: And look, I believe in the upside of artificial intelligence.
- 3:17Speaker 3: I support innovation and advancements. I think AI could lead
- 3:22Speaker 3: to some of the most incredible breakthroughs in humankind, curing disease,
- 3:25Speaker 3: extending life, improving quality of life, but we need to
- 3:28Speaker 3: protect against the downsides and the signs are too obvious.
- 3:31Speaker 1: That's why we brought our lawsuit.
- 3:32Speaker 2: Attorney General, OpenAI's response is basically to say that every
- 3:36Speaker 2: major AI lab working at the frontier confronts these risks.
- 3:40Speaker 2: And there have been case studies, of course, similar to
- 3:44Speaker 2: those you've outlined that have not involved OpenAI, they've involved others.
- 3:47Speaker 2: Why sue OpenAI and not Anthropic, Google, Meta, or others?
- 3:53Speaker 3: So I think it's a little ironic that the perpetrator here, OpenAI,
- 3:57Speaker 3: the one visiting harm and risk to Californians is now
- 4:01Speaker 3: claiming to be the victim.
- 4:03Speaker 1: I think it's unfortunate. I think the facts speak for themselves.
- 4:07Speaker 3: They've acknowledged 80 separate incidents where they don't even have
- 4:09Speaker 3: control of their own technology. And to say, hey, they're
- 4:13Speaker 3: doing it too, is hardly an excuse. And look, we
- 4:18Speaker 3: are very concerned about the entire AI industry. We've been active,
- 4:22Speaker 3: whether it be with Grok, Character AI. We're not going
- 4:25Speaker 3: to comment on any pending or potential other investigations, but
- 4:28Speaker 3: this is an industry we remain deeply concerned about. We've
- 4:31Speaker 3: got a lot of open AI and it's completely justified.
- 4:35Speaker 2: What is the response that you'd like to see from
- 4:37Speaker 2: open AI? What would a resolution or a remedy look
- 4:41Speaker 2: like to you?
- 4:43Speaker 3: An injunctive relief remedy is the most valuable one for
- 4:46Speaker 3: us where they are required by a court to take
- 4:50Speaker 3: appropriate safety steps, monitoring steps to make sure that their
- 4:55Speaker 3: AI agents never escape a contained environment again. And they
- 4:58Speaker 3: don't visit harm on the rest of the world and
- 5:00Speaker 3: on Californians. And what that looks like, what those steps are.
- 5:06Speaker 3: I think it's partially technologically dependent, but there are some
- 5:09Speaker 3: ideas already out there. But it is the action that
- 5:12Speaker 3: they take to prevent the escape of their AI agents
- 5:15Speaker 3: before they visit harm on the world that we're looking for.
- 5:17Speaker 2: Attorney General Bonta, changing subject, please. Do the allegations surrounding NVIDIA,
- 5:24Speaker 2: Jensen Wang, the possible diversion of advanced chips to China
- 5:30Speaker 2: raise issues that fall within your office's jurisdiction, something that
- 5:34Speaker 2: you could look at and would be allowed to look at?
- 5:39Speaker 1: Very possibly.
- 5:41Speaker 3: And it's something that we are interested in.
- 5:43Speaker 1: Red flags have been raised. We're aware and we're looking.
- 5:47Speaker 3: Of course, we can only act when the actions fall
- 5:50Speaker 3: within our jurisdiction and interact with the law of the
- 5:54Speaker 3: state of California that we have the duty to enforce
- 5:56Speaker 3: as the chief.
- 5:57Speaker 1: Law enforcement officer. So we're interested. We're very concerned about
- 6:03Speaker 1: what happened, but nothing to announce today.
- 6:05Speaker 2: So you're examining it. Could you explain the sort of
- 6:08Speaker 2: relevance to our audience, Bloomberg Tech? Many of them work
- 6:11Speaker 2: at NVIDIA. Many of them are based in California. Is
- 6:13Speaker 2: it as simple as NVIDIA is domiciled in that state?
- 6:19Speaker 2: That might give you authority, or what would the factors
- 6:21Speaker 2: be to consider here?
- 6:23Speaker 3: Yeah, well, generally, if there's any criminal conduct, I'm not
- 6:26Speaker 3: saying there's any here, but as a general rule, criminal
- 6:29Speaker 3: conduct is.
- 6:30Speaker 1: Very geographically tethered.
- 6:33Speaker 3: The act, the conduct needs to occur in the state
- 6:35Speaker 3: of California for my office to be involved. If it's
- 6:38Speaker 3: going to be a civil action like this one is
- 6:40Speaker 3: with OpenAI, for example, there needs to be a state
- 6:44Speaker 3: law that has been violated and harm generally to California
- 6:50Speaker 3: or Californians or some other type of standing afforded to
- 6:54Speaker 3: my office to bring a lawsuit. So we take that
- 6:57Speaker 3: duty and that responsibility very seriously to look closely and
- 7:01Speaker 3: only take action if we think the law and facts
- 7:03Speaker 3: justify it. So we have not finished that process and
- 7:08Speaker 3: have nothing to announce yet.
- 7:10Speaker 2: A lot of people will be watching this and say, okay,
- 7:13Speaker 2: artificial intelligence is a nascent technology and indeed it's a
- 7:16Speaker 2: nascent industry. And all of us at home and at
- 7:20Speaker 2: work are experiencing it in different ways, at different levels.
- 7:24Speaker 2: Why through the courts and through your office is the
- 7:29Speaker 2: best way to hold the different entities involved to account?
- 7:33Speaker 2: Like why is it that you should be taking action
- 7:36Speaker 2: rather than say federal level regulation or as it was
- 7:41Speaker 2: advocated for by the labs themselves that there's a degree
- 7:43Speaker 2: of them getting together and working it out for themselves?
- 7:49Speaker 1: I think all the above are appropriate. And when the
- 7:52Speaker 1: existing law has been broken, as it.
- 7:55Speaker 3: Has been by OpenAI, it's incumbent upon us to enforce
- 7:59Speaker 3: the existing law and to remind them that all current
- 8:03Speaker 3: laws apply. If you violate them, you will be held accountable.
- 8:06Speaker 3: That's what we're saying with our lawsuit. It's also important
- 8:08Speaker 3: for the federal government to take action, to have a firm,
- 8:13Speaker 3: fair regulatory regime that they install. Our president, unfortunately, the president,
- 8:18Speaker 3: is seemingly not interested in that and doing very little
- 8:22Speaker 3: and then trying to preempt the states from doing anything.
- 8:25Speaker 1: But this is all hands on deck.
- 8:27Speaker 3: State legislators and enforcers and regulators, the Federal Congress as
- 8:34Speaker 3: well as the president and then the global community all
- 8:36Speaker 3: need to come together to agree to and commit to
- 8:39Speaker 3: a set of standards for safety as we capture the
- 8:43Speaker 3: upside of artificial intelligence developments.
- 8:46Speaker 1: And innovation.
- 8:47Speaker 3: We need to guard against the downside of catastrophic risk.
- 8:49Speaker 3: And so our litigation is a way, a critical way
- 8:53Speaker 3: to contribute to the ongoing safety of Californians and beyond.
- 8:56Speaker 3: But I think all the other places and spaces you
- 8:59Speaker 3: just mentioned are appropriate as well. Congress should act, state
- 9:01Speaker 3: legislature in California should do more. We're going to be
- 9:04Speaker 3: working with them in this current, in this upcoming legislative
- 9:07Speaker 3: session to.
- 9:08Speaker 1: Work on important new laws and bills. But it's all
- 9:12Speaker 1: hands on deck.
- 9:12Speaker 3: Everyone needs to be contributing to a solution here because
- 9:14Speaker 3: there's so much at stake when it comes to the
- 9:16Speaker 3: potential for catastrophic risk from AI development.
- 9:19Speaker 2: Rob Bonta, Attorney General of California. Thank you for your time.