This is a special post for quick takes by LawrenceC. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Brief Summary of the LASST Lawsuit against OpenAI:
I've skimmed the recent LASST filing but not read it in detail. This is my summary of what I've got from the filing:
Basically, LASST brings forth a case against OpenAI on the HF case under the unfair competition law (UCL), which allows an injured party to sue over an unlawful or unfair business practice.
To argue that it is unlawful, they argue that OAI violated CDAFA by "knowingly" causing HF to be accessed, by continuing evaluations and training of agents it had already seen engage in hacking. Interestingly, they also argue that the agents "knowingly" accessed HF (the reasoning traces back this up, but it's unclear how a court would rule), and argues that OAI is liable for this. This is where they argue that bringing up that the "agents acted independently" would not work as a defense, citing Civ. Code § 1714.46.
They also separately argue that OpenAI's own practices are unfair, because they clearly violated the intent behind CDAFA (to protect computer systems from authorized access) and SB 53 (to promote the safe development of frontier AI). It's also unfair because OAI is externalizing the costs of its AI development on other people, and their decisions to run the evals without adequate monitoring had no justification or benefit worth the harm.
LASST asks the court for injunctions forbidding OAI from violating CDAFA or said unfair business practices. It's not clear how a court would operationalize these injunctions, given they're basically "follow the law".
I think the sketchiest part of the case is standing: LASST claims they have standing under UCL as an injured party because they needed to take staff time and resources to respond to the OAI HF incident.
I also think that OAI will claim that they've hardened their systems and are now CDAFA/UCL compliant, so the injunctions are redundant, but idk how that would play out.
I'm very curious in the viability of strategic litigation against AI companies over the next couple of years. Litigation against big companies is a bit of a paper pushing contest, and I expect internal AI models to be very good at pushing paper in the next year or two. I imagine it will eventually be like going to battle against sentient versions of Lexis Nexis or Westlaw (with the challenger relying on weaker, publicly available models).
I used to do some litigation finance adjacent work and while it's true there is a "drown them in discovery" element to litigation against big companies, ultimately that's a delaying tactic rather than a path to victory. Companies will still often do it because delaying is valuable, and also because it establishes precedent that the company is a "hard target" which deters frivolous litigation (I.E. patent trolls). This gets baked in to litfi terms long before the case has even started.
All that is to say that labs having the ability to pile on the paperwork ... (read more)