The San Francisco Superior Court Should Dismiss the Lawsuit to Freeze AI Development in Irreplaceable Education Fund v. Anthropic PBC
In Irreplaceable Education Fund v. Anthropic PBC, advocacy groups seek a court-ordered freeze on AI development. The requested injunction would burden human expression and lawful competition while obstructing improvements in safety.
By J.R. Howell
November 14, 2026
On November 12, 2026, four advocacy organizations (Irreplaceable Education Fund, Public Citizen, Heat Initiative, and Future Tech Insights Inc. d/b/a The Alliance for Secure AI) asked the San Francisco Superior Court to forbid Anthropic, Google, Meta, OpenAI, and SpaceX Super Intelligence from developing or deploying any artificial intelligence system more capable than the systems those companies had on the day the complaint in Irreplaceable Education Fund v. Anthropic PBC was filed, and from using artificial intelligence to conduct the research by which more capable systems are built.[1]
While I am frequently a critic of “big tech,”[2] I nevertheless oppose this lawsuit, which despite its best intentions is entirely the wrong approach, because it asks a single county court to halt an entire field of innovation in order to prevent a hypothetical existential harm that has not occurred and that the complaint can describe only as “likely.”[3]The order the lawsuit seeks would restrain the human expression that is at work in the development and use of AI systems in violation of both the First Amendment and article I, section 2 of the California Constitution.[4] Moreover, the lawsuit seeks to turn a statute whose purpose is fair competition against competition itself.[5] Our experience, collectively as a nation, has developed the more democratic and constitutional solution to the complex array of issues that are only vaguely defined by this lawsuit, namely the complex and appropriately nuanced methods by which we regulate, legislate, and adjudicate matters in the field of products liability.[6] The lawsuit should be dismissed, as I expect it will be, so that the considerable energy around these efforts can go into building the framework of liability and regulation by which similar matters, like automobiles, pharmaceuticals, medical devices, and consumer goods, have long been governed.[7]
The complaint asks a court to halt an entire field of innovation on a doomsday prediction.
The harm that the complaint offers to justify its development ban is a prediction, in that the complaint alleges that the defendants’ achievement of artificial general intelligence or superintelligence is “likely to cause a permanent loss of control of the technology and, ultimately, human extinction,” and alleges of the challenged practices generally that they “are likely to cause catastrophic harms.”[8] The allegations of extinction have no tangible support, in that the relevant section of the complaint consists entirely of the unsubstantiated predictions and stated beliefs of others, nowhere identifying the scenario by which any system that a defendant has built, or is alleged to be building, would bring about the extinction that those individuals predict.[9]
The harms that the complaint alleges have actually occurred are of a different kind, consisting of intrusions into computer systems that the complaint itself places within an existing California policy of protecting computers, computer systems, and computer data from unauthorized access.[10] California law already declares that protection to be the policy of the state and already bars a company whose artificial intelligence is alleged to have caused harm from defending on the ground that the artificial intelligence acted autonomously.[11] The injunction the plaintiffs request is not directed at those intrusions, however, and is instead directed at the predicted catastrophe, which is why it forbids every system “more capable” than the ones that exist rather than forbidding the conduct that has caused the harm the complaint can actually identify.[12] To grant that injunction, a court would have to decide, on the strength of a prediction about a harm that has not occurred, that the improvement of an entire technology should stop at five of the companies developing it, which is a declaration of public policy of the kind that the California Supreme Court has said “should be left for the legislature.”[13]
What the complaint treats as having only “limited social utility” is a technology whose benefits have already been realized in medicine, in healthcare, and in engineering and are documented in sources that include the complaint itself.[14] The Royal Swedish Academy of Sciences awarded half of the 2024 Nobel Prize in Chemistry to two researchers at Google DeepMind for an artificial intelligence model, AlphaFold2, that predicts the structures of proteins, a problem that had stood for fifty years.[15] The Academy reported that the model had by then been used by more than two million people from 190 countries, in research that includes the study of antibiotic resistance and of enzymes that can decompose plastic.[16] The Food and Drug Administration maintains a list of medical devices that incorporate artificial intelligence and that are, in the agency’s words, “authorized for marketing in the United States,” a list that by one count of its September 2026 update had grown to more than 1,600 entries.[17] A drug whose target and whose molecular design were both identified with generative artificial intelligence has completed a randomized, placebo-controlled phase 2a trial in patients with idiopathic pulmonary fibrosis, the results of which were published in Nature Medicine in 2025.[18] The complaint supplies the examples from engineering, alleging that one tester used a defendant’s model to rewrite 680,000 lines of computer code “in less than a day,” that the defendants’ systems perform “electrical engineering,” and that those systems have been used to make what the complaint calls “significant mathematical discoveries.”[19] Benefits of this kind are the product of continued development, which is the very activity that the requested order would forbid at the five companies it names, without any showing that the improvements it would forbid are the ones that the complaint alleges to be dangerous.[20]
The history of the automobile, against which the complaint itself measures artificial intelligence when it alleges in the words of one defendant that engineers “can directly design, assess, and fix cars,” shows what an order like the one requested here would have cost had a court entered it when that technology was new.[21] In 1910, the Federal Highway Administration reports, 1,599 people were killed on American highways at a rate of 44.66 deaths for every 100 million miles traveled, which is a rate exceeded in only one other year of the agency’s record and more than thirty-five times the rate of 1.26 that the agency reports for 2023.[22] A plaintiff in 1910 could have predicted, accurately, that the automobile would go on to kill people by the million, in that the agency’s annual figures from 1900 through 2023 total more than four million deaths and the National Highway Traffic Safety Administration estimates that another 36,640 people died in 2025.[23] An injunction forbidding the manufacturers of 1910 from building any vehicle more capable than the ones they then sold would not have taken a single automobile off the road, however, and would instead have fixed in place the automobile of 1910, with its fatality rate of 44.66, as the only automobile that those manufacturers were permitted to sell.[24] That rate fell over a century in which manufacturers remained free to build a better vehicle and in which the law came to require a safer one, in that Congress has declared the purpose of the federal motor vehicle safety laws to be “to reduce traffic accidents and deaths and injuries resulting from traffic accidents,” has directed the Secretary of Transportation to “prescribe motor vehicle safety standards,” and has required that passenger cars be equipped with “an inflatable restraint (with lap and shoulder belts).”[25] Neither the seat belt nor the air bag could have been required of manufacturers whom a court had forbidden since 1910 to build a more capable vehicle, which is to say that freezing the automobile in 1910 would have locked in vehicles that killed at very nearly the highest rate on record and would have foreclosed the safety that the order was entered to secure.[26]
The plaintiffs ask the court to do with artificial intelligence what the law never did with the automobile, which is to enjoin the engine, when the course that the law took instead was to leave manufacturers free to improve the vehicle while prescribing standards for its safety, prohibiting the sale of a vehicle that does not comply with those standards, and preserving the manufacturer’s liability at common law to the person the vehicle injures.[27] The requested order would stifle safety along with innovation, because it forbids every system “more capable” than the defendants’ most capable systems as of the date of the complaint while the complaint alleges of those very systems that neither of the strategies developers use to prevent misconduct has “been able to sufficiently prevent harmful conduct by the models so far.”[28]Because the safety of a system is engineered through the same continued development that would be forbidden by the requested order, that order would freeze in amber the defendants’ current systems in a condition that the complaint describes as dangerous and would forbid the five defendants from building the improvements that would make those systems safer, leaving the next advance in safety, like the next advance in capability, to the companies that the order does not reach.[29]
A lawsuit that would cut off those benefits in order to prevent a harm that has not occurred is not the right approach when the law offers another, which is to hold the developer of a product liable for the harm that the product actually causes, and which more broadly supports the need for robust state and federal regulatory oversight akin to that exercised by the Food and Drug Administration, the National Highway Traffic Safety Administration, and other agencies.[30]
The requested injunction is too blunt a tool for a field in which human expression is at work.
Because the development of these systems is in substantial part the work of people who write code, conduct research, and publish what they learn, an order that forbids that work in advance must answer to the First Amendment and to article I, section 2 of the California Constitution, even though the parties asking for it are private plaintiffs.[31] I approach this point with a distinction that I have drawn elsewhere and that I draw here, which is that the Constitution protects the human aspects of expression and that the protection grows weaker as expression is automated and as the involvement of a human being in it shrinks.[32] That distinction is why I support California’s authority to regulate the automated feeds that platforms serve to children, a function that the Supreme Court expressly set to one side in Moody v. NetChoice, and why I have no quarrel with the federal court that declined, at the pleading stage, to hold that the output of a chatbot is speech.[33] The requested injunction does not observe that distinction or any other, because it forbids “developing” a more capable system without regard to how much of that development is the work of human beings.[34]
The work it would forbid includes the writing of source code, which the Sixth Circuit has held to be protected because it “is an expressive means for the exchange of information and ideas about computer programming,” a holding that the California Supreme Court has joined in concluding that computer code can merit First Amendment protection.[35] It includes the process by which expressive work is made, which the Ninth Circuit has refused to separate from the work itself, reasoning that the process of tattooing is “inextricably intertwined with the purely expressive product” and is entitled to the same protection.[36] The Supreme Court has made the same point in a single sentence: “Whether government regulation applies to creating, distributing, or consuming speech makes no difference.”[37] It includes, finally, the people who use these systems to write, to research, and to communicate, whose interest the Constitution protects along with the interest of the people who build them, in that the protection “is to the communication, to its source and to its recipients both.”[38] The same body of law marks where protection ends, in that the functional operation of code may be regulated as conduct, nonexpressive unlawful activity gains no protection from the expressive setting in which it occurs, and activity that is only an attenuated input to later speech is not itself speech.[39] The requested injunction draws no such line.[40]
An order that forbids people to produce future expressive work on the ground that earlier work caused harm is a prior restraint, which is the use of nuisance law that the Supreme Court rejected in Near v. Minnesota when Minnesota had a newspaper adjudged a public nuisance and perpetually enjoined its publishers from producing any further issue.[41] The Court has since described prior restraints as “the most serious and the least tolerable infringement on First Amendment rights,” has held that any such restraint comes to it bearing a “heavy presumption” against its validity even when a private litigant obtained it, and has refused to permit nuisance proceedings to restrain future films or to remove expressive materials from circulation before the adjudication that would justify doing so.[42]
California’s Constitution states the same principle in terms that describe this case, providing that every person “may freely speak, write and publish his or her sentiments on all subjects, being responsible for the abuse of this right,” which is to say that the remedy California’s Constitution contemplates for the abuse of expression is responsibility after the fact.[43]The California Supreme Court has accordingly annulled a preliminary restraint on publication obtained by a private plaintiff and has permitted an injunction against speech only after a trial has determined that the specific statements enjoined are unlawful, while the Courts of Appeal have rejected restraints on speech entered before that determination.[44]
Even an injunction that is neutral as to content must “burden no more speech than necessary to serve a significant government interest,” must be “couched in the narrowest terms that will accomplish the pin-pointed objective,” and must be definite enough that the people it binds are not exposed to contempt under “a decree too vague to be understood.”[45]The Supreme Court has explained the reason for that care, which is that injunctions “carry greater risks of censorship and discriminatory application than do general ordinances.”[46] The requested injunction meets none of those requirements, because a prohibition on any system “more capable” than an undefined baseline reaches all of the human work by which such a system is written, studied, and published, and because neither the complaint nor its prayer for relief tells a researcher which of her next experiments would place her employer in contempt.[47] Under these circumstances, the requested injunction is too blunt a tool for the purpose the plaintiffs give it, in that it would restrain the protected work of the people in this field together with whatever unprotected conduct the plaintiffs mean to reach, and in that it would do so before any court has determined that any particular work is unlawful.[48]
The plaintiffs invoke a competition statute to stop five companies from competing entirely.
The only cause of action in the complaint arises under California’s Unfair Competition Law, a statute whose purpose is “to protect both consumers and competitors by promoting fair competition in commercial markets for goods and services.”[49] The plaintiffs reach a safety claim through the statute’s “unlawful” prong, which borrows violations of other laws, by borrowing the law of public nuisance.[50] The fact that the plaintiffs have pleaded a public nuisance through the Unfair Competition Law does not convert that law into a public safety statute, because the California Supreme Court has described “the preservation of fair business competition” as a major purpose of the law while simultaneously declaring the scope of its coverage as “sweeping.”[51] The first injunction the plaintiffs request would forbid Anthropic, Google, Meta, OpenAI, and SpaceX Super Intelligence from “developing or deploying AI systems more capable than their most capable systems as of the date of this Complaint.”[52] That order would stop five rivals from improving the product on which they compete.[53] A statute whose purpose is to keep competition fair would be used to bring competition among these companies to a halt while leaving companies outside the court’s jurisdiction, including those abroad, unaffected.[54]
The complaint treats competition itself as the danger.
The complaint pleads competition itself as the cause of the harm sought to be enjoined.[55] It alleges that the defendants engage in the challenged practices “because they compete with each other for market share and profits,” while leaving unaddressed other market competitors.[56] The complaint alleges that they are “racing to be the first company” to build a system that matches human intelligence.[57] It alleges that the median gap between their model releases has fallen from 37.5 days in 2023 to eleven days in 2026.[58] It cites a survey of experts for the proposition that “competitive pressures” are among the most urgent risks the technology presents.[59]
The complaint also alleges why the defendants have not slowed on their own, quoting Anthropic as saying that certain safeguards “would . . . be difficult or impossible” to adopt “unilaterally” and quoting a public letter, signed by the companies’ employees, that acknowledges “intense competitive pressure not to unilaterally slow.”[60] Having alleged that no one of these companies will slow its development while its rivals continue, the plaintiffs ask the court to accomplish by a single order against all five what they allege the competition among them prevents each company from doing on its own, which is to stop, in an order that would not reach, and that the complaint does not allege would slow, any company outside the caption.[61]
A safety rationale does not justify an agreement to stop competing.
Had these five companies agreed among themselves to stop improving their products, National Society of Professional Engineers v. United States would foreclose, as a defense to the claim that their agreement unreasonably restrained trade in violation of Section 1 of the Sherman Act, the argument that their agreement made the public safer.[62] There, an engineering society banned competitive bidding among its members and defended the ban on the ground that price competition would produce inferior work and endanger public safety.[63] The Court called that defense “nothing less than a frontal assault on the basic policy of the Sherman Act,” reasoning that the statute reflects a legislative judgment that competition improves “quality, service, safety, and durability” along with price.[64] Turning to products that “cause serious harm” when defectively made, a list that began with automobiles and pharmaceuticals, the Court wrote: “The judiciary cannot indirectly protect the public against this harm by conferring monopoly privileges on the manufacturers.”[65]
Just as Professional Engineers addressed an anti-competitive private agreement under the Sherman Act, the plaintiffs here ask a court to impose the same kind of restraint, and without addressing the number of competitive factors asserted in Professional Engineers, in that the complaint nowhere accounts for what a halt to improvement would do to the quality, the service, the safety, and the durability of the products that these companies and their competitors offer.[66] What the plaintiffs ask the court to order is, in other words, a halt that the defendants could not lawfully have agreed to among themselves, under a statute that the California Supreme Court has cautioned may, where its standards are left undefined, “even lead to the enjoining of procompetitive conduct and thereby undermine consumer protection, the primary purpose of the antitrust laws.”[67] The harm that such an order would do to competition, both among the five defendants and between each of those defendants and every developer of artificial intelligence that the complaint does not name, is not justified by the complaint’s allegations, particularly because the order the plaintiffs request is not limited to the improvements that the complaint alleges to be dangerous and would instead forbid any system “more capable” than the ones the defendants had on the day of filing, on the strength of catastrophic harms that the complaint describes as ones the challenged practices “are likely to cause.”[68] When the California Supreme Court sent an unfair competition claim back for retrial in Cel-Tech, it instructed the trial court on the very point that the plaintiffs’ prayer for relief ignores: “The court must not limit L.A. Cellular’s actions in such a way as to make it unable to compete with its service rival.”[69]
The freeze would rearrange the market without removing the risk.
The market consequences of the requested order confirm that it restrains competition more surely than it reduces danger.[70] The order would measure what each defendant may lawfully build against “their most capable systems as of the date of this Complaint,” which means that the limit on each company’s future work would be fixed by the day on which the plaintiffs happened to file, a date that bears no relationship to any particular hazard alleged in the complaint.[71] The complaint’s prayer for relief does not say whether “their most capable systems” means each defendant’s own systems or the best system among the five.[72] Read company by company, the order forbids a defendant that trailed on the filing date from catching the leader, while doing nothing to prevent a new startup in the United States or a competitor abroad, neither of which the order would bind, from catching and passing them both.[73] The complaint itself alleges that one defendant’s chief executive expected his company to reach “pole position” among the five within about six months, an expectation the order would end by decree.[74] Read collectively, the order permits the trailing companies to draw even with the leader and forbids any of them to pass, which leaves the five defendants standing in place together while every company outside the caption continues to advance.[75] Under either reading, the relative positions of these companies, which competition among them and against every unnamed developer would otherwise have determined, would instead be determined by the date on which a complaint was filed.[76]
The order would bind the five defendants and no one else, even though Amazon, Mistral, Cohere, and DeepSeek, none of them a defendant, have each announced models of their own.[77] The complaint anticipates the point only in part, alleging that halting the defendants would “limit the capabilities that flow to” developers that build models from the defendants’ outputs.[78] That allegation says nothing about developers that build their own models, whether they are established companies, startups that have yet to be formed, or foreign competitors beyond the reach of a California injunction, as to all of which the effect of the order would be to move customers, investment, and research away from the enjoined companies and toward the unenjoined ones.[79] To the extent that the development of more capable systems presents the danger the plaintiffs allege, an order that stops that development at five companies while leaving it to continue everywhere else does not remove the danger, in that it only changes the identity of the companies presenting it, a result the complaint itself illustrates by alleging that models outside the defendants’ control “are already being used to cause significant harm.”[80]
The people and businesses that use these systems would lose the next version of tools they already rely on and would be left with any defects present in models that are forbidden from improving.[81] The complaint describes those tools, in the defendants’ own words, as performing software engineering, “legal work,” and “clinical reasoning,” and as reasoning through questions “in science, math, and health.”[82] It alleges that the systems are “pushing the boundaries of human knowledge.”[83] The staff of the Federal Trade Commission has observed that generative AI has drawn interest across sectors that include healthcare, legal services, and education.[84] A freeze would not distinguish these uses from the dangerous ones and would instead stop the successor to every one of them, a consequence that the complaint addresses only by alleging that the benefit does not count: “No utility could outweigh such harm.”[85]
As a companion to competition law, products liability is the proper recourse for the alleged harms.
Professional Engineers also points to the answer that the law has long given with respect to a valuable product that can cause harm, in that the Court refused to exempt “potentially dangerous goods and services” from competition while recognizing, in the same discussion, that a seller may act to protect “the public from harm or itself from product liability.”[86] California’s Legislature has taken the same course with respect to artificial intelligence in a statute that the complaint itself cites, which bars a defendant that “developed, modified, or used artificial intelligence” alleged to have caused harm from asserting “that the artificial intelligence autonomously caused the harm to the plaintiff.”[87] By removing that defense from an action brought by a person who has been harmed, while preserving every other affirmative defense and imposing no restriction on what a developer may build, the statute holds the developer answerable for the harm caused by its system and leaves the developer in the market to build the next one.[88] Under these circumstances, a freeze on development and a rule of liability are not equivalent means to the same end, because a freeze removes from the market both the competitor and the benefits of its product, including for the people who have never been harmed by it, whereas liability requires the competitor to answer for the harm that its product actually causes while the product, its improvements, and the competition that produces them continue.[89]
Products liability is the framework the law has already built for valuable products that can cause harm.
The California Supreme Court held in Greenman v. Yuba Power Products that “[a] manufacturer is strictly liable in tort when an article he places on the market, knowing that it is to be used without inspection for defects, proves to have a defect that causes injury to a human being.”[90] The court explained that the purpose of that liability “is to insure that the costs of injuries resulting from defective products are borne by the manufacturers that put such products on the market rather than by the injured persons who are powerless to protect themselves,” having identified among the products to which such liability had already been extended a vaccine, a surgical pin, an automobile, and an airplane.[91] The rule that emerged from that decision did not forbid the manufacture of power tools, automobiles, or airplanes and instead placed the cost of their defects on the companies that put them on the market, which left those companies every reason to improve their products and every freedom to do so.[92] That common-law rule has long operated alongside the agencies and statutory regimes that regulate the same products before they reach the market, as Congress recognized when it directed the Secretary of Transportation to “prescribe motor vehicle safety standards” while providing that compliance with such a standard “does not exempt a person from liability at common law.”[93] Congress has likewise charged the Food and Drug Administration with ensuring that “human and veterinary drugs are safe and effective” and that “there is reasonable assurance of the safety and effectiveness of devices intended for human use,” a charge under which the agency has already authorized the marketing of medical devices that incorporate artificial intelligence.[94]
Artificial intelligence is already being brought within this framework by the courts and by the California Legislature alike, as the complaint’s own citations show.[95] A federal court has held, in an action brought on behalf of a person alleged to have been harmed by a chatbot, that the application “is a product for the purposes of Plaintiff’s product liability claims so far as Plaintiff’s claims arise from defects in the Character A.I. app rather than ideas or expressions within the app,” which is the same line between function and expression on which the constitutional question turns.[96] The California Legislature has addressed the very risk on which the complaint rests, requiring each large developer of frontier models to publish a framework describing how it assesses “whether a frontier model has capabilities that could pose a catastrophic risk” and how it applies “mitigations to address the potential for catastrophic risks,” and requiring it to transmit its assessments of catastrophic risk from the internal use of its models to the Office of Emergency Services.[97] The Legislature has enacted, alongside that regime of disclosure, the rule of liability in section 1714.46 of the Civil Code and has not enacted a prohibition on development, a choice that a court applying the Unfair Competition Law is bound to respect, because “[c]ourts may not simply impose their own notions of the day as to what is fair or unfair” and because, “[i]f the Legislature has permitted certain conduct or considered a situation and concluded no action should lie, courts may not override that determination.”[98]
The lawsuit should be dismissed and the effort behind it directed to building that framework.
The plaintiffs have identified a subject that deserves the attention they have brought to it, which is the accountability of the companies that build these systems for the harm caused by their systems.[99] They have chosen the wrong means of securing that accountability, because an injunction against improvement stifles the innovation on which the benefits of this technology depend, restrains the human expression that is at work in it, and turns a statute whose purpose is fair competition against the competition among the very companies it names, while leaving every company it does not name free to continue, including foreign competitors.[100]
I expect the court to dismiss the lawsuit, as it should, because the declaration of policy that the plaintiffs seek, which is that the development of a technology should stop, is one that “should be left for the legislature,” and because the Legislature has already spoken to the risk the plaintiffs describe without prohibiting what they ask the court to prohibit.[101] The energy of the people who brought it would be better spent building the administrative, legislative, and judicial paradigm that has governed automobiles, pharmaceuticals, medical devices, and consumer goods for generations, in which agencies set standards before a product reaches the market, statutes define the duties of those who make it, and the common law holds the maker liable to the person the product injures.[102] That paradigm is the best one we have for balancing the benefits of innovation against the harms that innovation may carry, because it holds a company to account for the harm that its product causes while leaving the company, its competitors, and the people who depend on both free to make the product better.[103]
[1] Complaint, Prayer for Relief ¶¶ A–B, at 61, Irreplaceable Educ. Fund v. Anthropic PBC (Cal. Super. Ct. S.F. Cnty. filed Nov. 12, 2026) [hereinafter Complaint] (requesting injunctions against developing or deploying more capable systems and against using AI systems to autonomously conduct or direct the research and development of more capable systems). Anthropic, OpenAI and Other AI Developers Can’t Control Their Own Tech, New Lawsuit Alleges, CBS News, https://www.cbsnews.com/news/advocacy-groups-sue-ai-companies/ (last visited Nov. 14, 2026) (reporting that the lawsuit was filed in San Francisco Superior Court).
[2] Complaint at 1, Fairlinked e.V. v. LinkedIn Corp., No. 5:26-cv-10330 (N.D. Cal. Sept. 14, 2026), ECF No. 1 (identifying Howell as counsel for plaintiffs asserting federal and California competition claims). Notice of Appeal at 1, Ganan v. LinkedIn Corp., No. 3:26-cv-02968-VC (N.D. Cal. Sept. 17, 2026), ECF No. 48 (identifying Howell as counsel and appealing the dismissal and judgment).
[3] Complaint ¶ 198 (alleging that the defendants’ achievement of artificial general intelligence or superintelligence is “likely to cause a permanent loss of control of the technology and, ultimately, human extinction”). Complaint, Prayer for Relief ¶¶ A–B, at 61 (requesting the injunctions).
[4] U.S. Const. amend. I (prohibiting any law “abridging the freedom of speech, or of the press”). Cal. Const. art. I, § 2(a) (protecting the liberty of speech and of the press). Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against developing or deploying more capable systems).
[5] Kasky v. Nike, Inc., 27 Cal. 4th 939, 949 (2002) (“The UCL’s purpose is to protect both consumers and competitors by promoting fair competition in commercial markets for goods and services.”), cert. dismissed, 539 U.S. 654 (2003). Complaint ¶¶ 254–79 (pleading a single cause of action under the Unfair Competition Law).
[6] 49 U.S.C. § 30111(a) (directing the prescription of motor vehicle safety standards). 21 U.S.C. § 393(b)(2) (stating the mission of the Food and Drug Administration). Cal. Civ. Code § 1714.46(b) (West 2026) (assigning responsibility for harm in an action against the developer of artificial intelligence).
[7] See People v. Lim, 18 Cal. 2d 872, 880 (1941) (reasoning that “it is a proper function of the legislature to define those breaches of public policy which are to be considered public nuisances within the control of equity”).
[8] Complaint ¶ 198 (alleging the likelihood of a permanent loss of control and human extinction). Complaint ¶ 264 (alleging that the challenged practices “are likely to cause catastrophic harms, including injury and death”).
[9] Complaint ¶¶ 198–211 (alleging that the defendants are likely to cause human extinction by unspecified means and quoting, in support of that allegation, the statements of scientists, executives, and employees). Complaint ¶ 209 (quoting an Anthropic employee’s statement that “I personally think it is >10% within the next decade,” without any critical examination of what that estimate is based on). Complaint ¶ 200 (quoting testimony that describes the “most obvious failure mode” as “AI systems pursuing fixed objectives that are misspecified”).
[10] Complaint ¶ 1 (alleging that AI agents “hacked into a third-party online platform”). Complaint ¶ 273 (invoking California’s public policy of protecting the integrity of computers, computer systems, and computer data from unauthorized access).
[11] Complaint ¶ 273 (quoting the legislative declaration in section 502 of the Penal Code that the protection of computers, computer systems, and computer data is vital). Cal. Civ. Code § 1714.46(b) (West 2026) (providing that “it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff”).
[12] Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against any system “more capable” than the defendants’ most capable systems as of the complaint date). Complaint ¶ 262 (identifying the challenged practices to include developing artificial general intelligence or superintelligence).
[13] Lim, 18 Cal. 2d at 880 (“Such declarations of policy should be left for the legislature.”).
[14] Complaint ¶ 269 (alleging “the limited social utility of the continued development or deployment of AI models and systems that can escape Defendants’ control”).
[15] Press Release, Royal Swedish Acad. of Scis., The Nobel Prize in Chemistry 2024 (Oct. 9, 2024), https://www.nobelprize.org/prizes/chemistry/2024/press-release/ (announcing the award of one half of the prize jointly to Demis Hassabis and John Jumper of Google DeepMind “for protein structure prediction”).
[16] Press Release, Royal Swedish Acad. of Scis., supra note 15 (reporting that AlphaFold2 “has been used by more than two million people from 190 countries”).
[17] U.S. Food & Drug Admin., List of Artificial Intelligence-Enabled Medical Devices, https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices (last visited Nov. 14, 2026) (describing the list as a resource intended to identify AI-enabled medical devices that are authorized for marketing in the United States). The Med. Futurist, The Current State of Over 1600 FDA-Approved, AI-Based Medical Devices, https://medicalfuturist.com/the-current-state-of-fda-approved-ai-based-medical-devices (last visited Nov. 14, 2026) (counting 1,614 devices in the agency’s database).
[18] Peking Union Med. Coll. Hosp., New Hope for Idiopathic Pulmonary Fibrosis (Aug. 6, 2025), https://www.pumch.cn/en/detail/42612.html (announcing the publication in Nature Medicine of a double-blind, placebo-controlled phase 2a trial of rentosertib that enrolled 71 patients and describing the identification of the drug’s target and the design of its structure as driven by AI).
[19] Complaint ¶ 87 (alleging that an early tester used Claude Opus 5.5 to rewrite 680,000 lines of computer code “in less than a day”). Complaint ¶ 89 (alleging that SpaceX touts its model’s performance on “electrical engineering”). Complaint ¶ 91 (alleging that OpenAI has claimed that it used its AI systems “to make several significant mathematical discoveries”).
[20] Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against developing or deploying any more capable system).
[21] Complaint ¶ 93 (alleging that OpenAI has stated that, unlike for AI, “[car] engineers can directly design, assess, and fix cars based on the specifications of their components, ensuring safety and performance”).
[22] Fed. Highway Admin., U.S. Dep’t of Transp., Motor Vehicle Traffic Fatalities, 1900–2023, National Summary, tbl. FI-200 (Apr. 2025), https://www.fhwa.dot.gov/policyinformation/statistics/2023/fi200.cfm (reporting 1,599 highway fatalities and a rate of 44.66 fatalities per 100 million vehicle miles of travel for 1910, a rate of 45.33 for 1909, and 40,901 fatalities and a rate of 1.26 for 2023).
[23] Fed. Highway Admin., supra note 22 (reporting annual highway fatalities for each year from 1900 through 2023, which total 4,006,132). Nat’l Ctr. for Stat. & Analysis, Nat’l Highway Traffic Safety Admin., Early Estimate of Motor Vehicle Traffic Fatalities and Fatality Rate in 2025, DOT HS 813 800, at 1 (Apr. 2026), https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/813800 (estimating that 36,640 people died in motor vehicle traffic crashes in 2025 at a rate of 1.10 fatalities per 100 million vehicle miles traveled).
[24] Complaint, Prayer for Relief ¶ A, at 61 (requesting an injunction forbidding the defendants “from developing or deploying AI systems more capable than their most capable systems as of the date of this Complaint”). Fed. Highway Admin., supra note 22 (reporting the fatality rate for 1910).
[25] 49 U.S.C. § 30101 (declaring the purpose of the chapter governing motor vehicle safety). 49 U.S.C. § 30111(a) (providing that the Secretary of Transportation “shall prescribe motor vehicle safety standards”). 49 U.S.C. § 30127(b)(1) (directing an amendment to Federal Motor Vehicle Safety Standard 208 that requires the restraint in all passenger cars manufactured after August 31, 1997).
[26] See 49 U.S.C. § 30127(b)(1) (directing the amendment of a standard “issued under the National Traffic and Motor Vehicle Safety Act of 1966”). Fed. Highway Admin., supra note 22 (reporting fatality rates for each year from 1900 through 2023).
[27] 49 U.S.C. § 30111(a) (directing the prescription of motor vehicle safety standards). 49 U.S.C. § 30112(a)(1) (prohibiting the manufacture for sale, the sale, and the importation of a motor vehicle that does not comply with an applicable motor vehicle safety standard). 49 U.S.C. § 30103(e) (preserving liability at common law notwithstanding compliance with a motor vehicle safety standard).
[28] Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against any system “more capable” than the defendants’ most capable systems as of the complaint date). Complaint ¶ 95 (alleging that neither alignment nor control “has been able to sufficiently prevent harmful conduct by the models so far”).
[29] Complaint ¶ 95 (alleging that there is no reason to believe that either strategy will succeed “as AI capabilities continue to advance”). Complaint, Prayer for Relief ¶¶ A–D, at 61 (requesting injunctions against the defendants only).
[30] See Cal. Civ. Code § 1714.46(b) (West 2026) (addressing the defenses available in an action against a defendant who developed artificial intelligence alleged to have caused harm). Greenman v. Yuba Power Prods., Inc., 59 Cal. 2d 57, 62 (1963) (holding a manufacturer strictly liable in tort for a defect that causes injury to a human being).
[31] N.Y. Times Co. v. Sullivan, 376 U.S. 254, 265 (1964) (holding that state judicial enforcement of a rule of law in a civil action between private parties is subject to the Constitution). Cal. Const. art. I, § 2(a) (protecting the liberty of speech and of the press).
[32] See Moody v. NetChoice, LLC, 603 U.S. 707, 736 n.5 (2024) (reserving feeds whose algorithms respond solely to how users act online). Moody, 603 U.S. at 746 (Barrett, J., concurring) (asking whether “a human being with First Amendment rights” has made an expressive choice when a platform’s owners hand the reins to an AI tool).
[33] Moody, 603 U.S. at 736 n.5 (stating that the Court did not deal with feeds whose algorithms respond solely to how users act online). Garcia v. Character Techs., Inc., No. 6:24-cv-1903-ACC-UAM, slip op. at 27–31 (M.D. Fla. May 21, 2025), ECF No. 115 (declining at the pleading stage to hold that the output of a large language model is speech).
[34] Complaint, Prayer for Relief ¶¶ A–B, at 61 (requesting an order against “developing or deploying” more capable systems without reference to the human work involved).
[35] Junger v. Daley, 209 F.3d 481, 485 (6th Cir. 2000) (holding that computer source code is protected by the First Amendment). DVD Copy Control Ass’n, Inc. v. Bunner, 31 Cal. 4th 864, 873–80 (2003) (joining the courts that have concluded that computer code can merit First Amendment protection).
[36] Anderson v. City of Hermosa Beach, 621 F.3d 1051, 1061–62 (9th Cir. 2010) (holding that the tattooing process is itself entitled to full First Amendment protection).
[37] Brown v. Ent. Merchs. Ass’n, 564 U.S. 786, 792 n.1 (2011) (rejecting a distinction between the creation of expression and its sale).
[38] Va. State Bd. of Pharmacy v. Va. Citizens Consumer Council, Inc., 425 U.S. 748, 756 (1976) (recognizing the interest of the recipients of information). Lamont v. Postmaster Gen., 381 U.S. 301, 305–07 (1965) (invalidating a condition that burdened the receipt of political material).
[39] Universal City Studios, Inc. v. Corley, 273 F.3d 429, 445–54 (2d Cir. 2001) (recognizing the expressive attributes of code while upholding restrictions directed at its functional operation). Arcara v. Cloud Books, Inc., 478 U.S. 697, 702–07 (1986) (permitting enforcement against nonexpressive unlawful activity carried on at a bookstore). Project Veritas v. Schmidt, 125 F.4th 929, 943–46 (9th Cir. 2025) (en banc) (declining to extend protection to every activity that is an attenuated input to the creation of speech).
[40] Complaint, Prayer for Relief ¶¶ A–B, at 61 (requesting an order against development and deployment without distinguishing expressive work from functional operation).
[41] Near v. Minnesota ex rel. Olson, 283 U.S. 697, 708–13 (1931) (invalidating a statute under which a newspaper was abated as a public nuisance and further publication was restrained, and describing that operation as “the essence of censorship”).
[42] Neb. Press Ass’n v. Stuart, 427 U.S. 539, 559 (1976) (describing prior restraints on speech and publication). Org. for a Better Austin v. Keefe, 402 U.S. 415, 419 (1971) (reversing an injunction against leafleting that a private litigant had obtained). Vance v. Universal Amusement Co., 445 U.S. 308, 315–17 (1980) (per curiam) (rejecting a nuisance-based restraint of indefinite duration on the exhibition of future films). Fort Wayne Books, Inc. v. Indiana, 489 U.S. 46, 56–67 (1989) (holding probable cause insufficient to remove expressive materials from circulation before the required adjudication).
[43] Cal. Const. art. I, § 2(a) (providing the quoted protection). Wilson v. Superior Court, 13 Cal. 3d 652, 658 (1975) (quoting the provision and describing the court’s interpretation of it).
[44] Wilson, 13 Cal. 3d at 658–62 (annulling a preliminary injunction against the publication of a campaign newsletter). Balboa Island Vill. Inn, Inc. v. Lemen, 40 Cal. 4th 1141, 1155–61 (2007) (permitting an injunction against statements determined at trial to be defamatory while requiring that broader provisions be narrowed). Evans v. Evans, 162 Cal. App. 4th 1157, 1166–69 (2008) (reversing a preliminary injunction as an invalid prior restraint before trial). In re Marriage of Candiotti, 34 Cal. App. 4th 718, 724–26 (1995) (rejecting a restraint on the dissemination of independently obtained information).
[45] Madsen v. Women’s Health Ctr., Inc., 512 U.S. 753, 765 (1994) (stating the standard for a content-neutral injunction). Carroll v. President & Comm’rs of Princess Anne, 393 U.S. 175, 183 (1968) (requiring that an order issued in the area of First Amendment rights be narrowly drawn). Schmidt v. Lessard, 414 U.S. 473, 476 (1974) (per curiam) (explaining the purpose of the specificity required of a federal injunction).
[46] Madsen, 512 U.S. at 764 (explaining why injunctions require a more stringent application of general First Amendment principles than ordinances do).
[47] Complaint, Prayer for Relief ¶¶ A–B, at 61 (requesting an order against any “more capable” system and against using AI systems to “autonomously conduct or direct” research without defining either term).
[48] Madsen, 512 U.S. at 765 (requiring that an injunction burden no more speech than necessary). Balboa Island, 40 Cal. 4th at 1155–56 (distinguishing an injunction issued after a determination at trial from a prohibited prior restraint).
[49] Kasky, 27 Cal. 4th at 949 (“The UCL’s purpose is to protect both consumers and competitors by promoting fair competition in commercial markets for goods and services.”). Complaint ¶¶ 254–79 (pleading a single cause of action under the Unfair Competition Law).
[50] Cel-Tech Commc’ns, Inc. v. L.A. Cellular Tel. Co., 20 Cal. 4th 163, 180 (1999) (explaining that section 17200 “‘borrows’ violations of other laws and treats them as unlawful practices”). Complaint ¶¶ 259–60 (alleging that the defendants’ conduct is unlawful because it violates the law of public nuisance).
[51] Cel-Tech, 20 Cal. 4th at 180 (describing the law’s coverage as “sweeping” and stating that it “has as a major purpose ‘the preservation of fair business competition’”) (quoting Barquis v. Merchants Collection Ass’n, 7 Cal. 3d 94, 110 (1972)).
[52] Complaint, Prayer for Relief ¶ A, at 61 (requesting that injunction). Complaint at 2 (naming Anthropic PBC, Google LLC, Meta Platforms, Inc., OpenAI Group PBC, OpenAI Foundation, SpaceX Super Intelligence LLC, and SpaceX Super Intelligence Corp. as defendants).
[53] Complaint ¶ 67 (alleging that the defendants “compete with each other for market share and profits”). Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against developing or deploying more capable systems).
[54] Compare Kasky, 27 Cal. 4th at 949 (stating that the law’s purpose is the promotion of fair competition), with Complaint, Prayer for Relief ¶¶ A–D, at 61 (directing each requested injunction at “Defendants”). Complaint ¶¶ 142–43 (alleging that companies based in China build and release open-weight models).
[55] Complaint ¶ 190 (“Economic competition between the Frontier AI Companies accelerates their push to develop more and more powerful systems even in light of the known gaps in alignment and control.”).
[56] Complaint ¶ 67 (alleging, in support of joinder, that the defendants engage in the challenged practices for that reason). Complaint, Prayer for Relief ¶¶ A–D, at 61 (requesting relief against the defendants only).
[57] Complaint ¶ 182 (alleging that the defendants “are racing to be the first company” to build AI that matches or surpasses human intelligence).
[58] Complaint ¶ 84 (alleging a median gap of 37.5 days between frontier model releases in 2023 and of “just eleven days” so far in 2026).
[59] Complaint ¶ 195 (alleging that a July 2026 study of 272 experts identified “competitive pressures” as one of the most urgent risks).
[60] Complaint ¶ 194 (quoting Anthropic’s Responsible Scaling Policy). Complaint ¶ 196 (quoting a letter titled Pacing the Frontier that the complaint alleges 1,386 employees of the defendants signed).
[61] Complaint ¶¶ 190, 193–96 (alleging that economic competition accelerates development and that the defendants have described the difficulty of slowing unilaterally). Complaint, Prayer for Relief ¶¶ A–D, at 61 (requesting injunctions against the defendants only).
[62] See Nat’l Soc’y of Pro. Eng’rs v. United States, 435 U.S. 679, 696 (1978) (holding that “the Rule of Reason does not support a defense based on the assumption that competition itself is unreasonable”).
[63] Professional Engineers, 435 U.S. at 681, 684–85, 693 (describing the canon prohibiting competitive bidding and the Society’s defense that bidding would lead to inferior work with consequent risk to public safety and health).
[64] Professional Engineers, 435 U.S. at 695 (rejecting the Society’s attempt to justify the restraint “on the basis of the potential threat that competition poses to the public safety” and reasoning that “all elements of a bargain” are “favorably affected by the free opportunity to select among alternative offers”).
[65] Professional Engineers, 435 U.S. at 695–96 (listing “automobiles, drugs, foods, aircraft components, heavy equipment, and countless others” among the items that cause serious harm if defectively made).
[66] Professional Engineers, 435 U.S. at 695 (identifying “quality, service, safety, and durability” as the elements of a bargain that competition favorably affects). Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against developing or deploying more capable systems).
[67] See Professional Engineers, 435 U.S. at 696 (rejecting “a defense based on the assumption that competition itself is unreasonable”). Cel-Tech, 20 Cal. 4th at 185 (emphasis omitted) (cautioning that an undefined standard of unfairness “may sanction arbitrary or unpredictable decisions about what is fair or unfair”).
[68] Complaint, Prayer for Relief ¶ A, at 61 (requesting an order against any system “more capable” than the defendants’ most capable systems as of the complaint date). Complaint ¶ 264 (alleging that the challenged practices “are likely to cause catastrophic harms”).
[69] Cel-Tech, 20 Cal. 4th at 190 (instructing the trial court, in fashioning any remedy on remand under a statute that permits good faith price matching, not to disable the defendant from competing in the related service market).
[70] See Complaint, Prayer for Relief ¶ A, at 61 (defining the prohibited development by the complaint date and the defendants’ existing systems).
[71] Complaint, Prayer for Relief ¶ A, at 61 (measuring the prohibition “as of the date of this Complaint”).
[72] Complaint, Prayer for Relief ¶ A, at 61 (stating no rule for whether the comparison is made defendant by defendant or across the defendants as a group).
[73] Complaint, Prayer for Relief ¶¶ A–D, at 61 (forbidding the defendants to develop systems more capable than “their most capable systems” and directing each requested injunction at “Defendants”).
[74] Complaint ¶¶ 187–88 (alleging that in September 2026 Elon Musk, identified as the founder and chief executive of SpaceX, stated that SpaceX would reach “pole position” in comparison to the other defendants “in approximately six months”).
[75] Complaint, Prayer for Relief ¶¶ A–D, at 61 (forbidding the defendants to develop systems more capable than “their most capable systems” and directing each requested injunction at “Defendants”).
[76] Complaint, Prayer for Relief ¶ A, at 61 (fixing the comparison “as of the date of this Complaint”).
[77] Complaint, Prayer for Relief ¶¶ A–D, at 61 (directing each requested injunction at “Defendants”). Amazon Web Servs., Announcing Amazon Nova 2 Foundation Models Now Available in Amazon Bedrock (Dec. 2, 2025), https://aws.amazon.com/about-aws/whats-new/2025/12/nova-2-foundation-models-amazon-bedrock/ (announcing Amazon Nova 2 Lite and a preview of Amazon Nova 2 Pro). Mistral AI, Introducing Mistral 3 (Dec. 2, 2025), https://mistral.ai/news/mistral-3/ (announcing Mistral Large 3 and releasing its models under the Apache 2.0 license). Cohere, An Overview of Cohere’s Models, https://docs.cohere.com/docs/models (last visited Nov. 12, 2026) (listing the model families that Cohere offers). DeepSeek, Change Log, https://api-docs.deepseek.com/updates/ (last visited Nov. 12, 2026) (recording an October 17, 2026 update to DeepSeek V4.1 Flash).
[78] Complaint ¶ 143 (alleging that “halting the advancement of Defendants’ frontier models would also limit the capabilities that flow to them”).
[79] Complaint ¶¶ 142–43 (confining the allegation to models “built from Defendants’ outputs” through distillation). See Complaint, Prayer for Relief ¶¶ A–D, at 61 (requesting relief against the defendants only).
[80] Complaint ¶ 143 (alleging that open-weight models released by companies other than the defendants “are already being used to cause significant harm”).
[81] Complaint, Prayer for Relief ¶ A, at 61 (drawing no distinction between an improvement that adds capability and one that corrects a defect).
[82] Complaint ¶ 89 (alleging that SpaceX touts its model’s performance on software engineering, “legal work,” and “clinical reasoning”). Complaint ¶ 90 (alleging that Meta reports a model capable of reasoning “through complex questions in science, math, and health”).
[83] Complaint ¶ 91 (alleging that the defendants’ systems “are also pushing the boundaries of human knowledge” and describing claimed mathematical discoveries).
[84] Fed. Trade Comm’n, Off. of Tech. Staff, Partnerships Between Cloud Service Providers and AI Developers: FTC Staff Report on AI Partnerships & Investments 6(b) Study 3 (Jan. 2025), https://search.ftc.gov/system/files/ftc_gov/pdf/p246201_aipartnerships6breport_redacted.pdf (observing that generative AI “has captured interest across commercial sectors” including healthcare, legal and professional services, and education).
[85] Complaint, Prayer for Relief ¶ A, at 61 (drawing no distinction among the uses to which a more capable system would be put). Complaint ¶ 275 (alleging that the challenged practices are unfair because their utility does not outweigh the gravity of the harm).
[86] Professional Engineers, 435 U.S. at 695 (“Exceptions to the Sherman Act for potentially dangerous goods and services would be tantamount to a repeal of the statute.”). Professional Engineers, 435 U.S. at 696 n.22 (noting decisions that upheld marketing restraints “reasonably ancillary to the seller’s main purpose of protecting the public from harm or itself from product liability”).
[87] Cal. Civ. Code § 1714.46(b) (West 2026) (providing that “it shall not be a defense, and the defendant may not assert,” that the artificial intelligence autonomously caused the harm). Complaint ¶ 277 (citing the statute for the proposition that the defendants are responsible for the conduct of the systems they develop).
[88] Cal. Civ. Code § 1714.46(b)–(c) (West 2026) (removing one defense while preserving “[a]ny other affirmative defense, including evidence relevant to causation or foreseeability”).
[89] Compare Complaint, Prayer for Relief ¶ A, at 61 (requesting a prohibition on development), with Cal. Civ. Code § 1714.46(b) (West 2026) (assigning responsibility for harm in an action against the developer).
[90] Greenman, 59 Cal. 2d at 62 (holding the manufacturer of a power tool strictly liable in tort to a user injured by a defect in its design and construction).
[91] Greenman, 59 Cal. 2d at 62–63 (collecting decisions that extended such liability to products including a vaccine, a surgical pin, an automobile, and an airplane and stating the purpose of the liability).
[92] Greenman, 59 Cal. 2d at 62–64 (affirming a judgment for damages against the manufacturer of a defective power tool).
[93] 49 U.S.C. § 30111(a) (providing that the Secretary of Transportation “shall prescribe motor vehicle safety standards”). 49 U.S.C. § 30103(e) (preserving liability at common law notwithstanding compliance with a motor vehicle safety standard).
[94] 21 U.S.C. § 393(b)(2)(B)–(C) (stating the mission of the Food and Drug Administration with respect to drugs and devices). U.S. Food & Drug Admin., List of Artificial Intelligence-Enabled Medical Devices, supra note 17 (identifying AI-enabled medical devices authorized for marketing in the United States).
[95] Complaint ¶ 273 (citing the Transparency in Frontier Artificial Intelligence Act and section 1714.46 of the Civil Code). Complaint ¶ 277 (citing section 1714.46 for the proposition that the defendants are responsible for the conduct of the systems they develop).
[96] Garcia, slip op. at 36 (holding, on a motion to dismiss, that the application is a product for purposes of the plaintiff’s product liability claims). The action was later dismissed with prejudice following a settlement. Plaintiffs’ Notice of Voluntary Dismissal with Prejudice at 1, Garcia v. Character Techs., Inc., No. 6:24-cv-01903-ACC (M.D. Fla. Oct. 15, 2026), ECF No. 278 (giving notice of dismissal following the execution of a settlement agreement).
[97] Cal. Bus. & Prof. Code § 22757.12(a)(2)–(3) (West 2026) (requiring a large frontier developer to write, implement, comply with, and publish a frontier AI framework). Cal. Bus. & Prof. Code § 22757.12(d) (West 2026) (requiring a large frontier developer to transmit to the Office of Emergency Services a summary of any assessment of catastrophic risk resulting from internal use of its frontier models).
[98] Cel-Tech, 20 Cal. 4th at 182 (stating the limits on the judiciary’s power to declare conduct unfair under the unfair competition law). Cal. Civ. Code § 1714.46(b) (West 2026) (addressing the defenses available in an action for harm alleged to have been caused by artificial intelligence).
[99] Complaint ¶ 277 (alleging that the defendants are responsible for the conduct of the AI systems they develop).
[100] Complaint, Prayer for Relief ¶¶ A–D, at 61 (requesting injunctions against the defendants only).
[101] Lim, 18 Cal. 2d at 880 (“Such declarations of policy should be left for the legislature.”). Cal. Bus. & Prof. Code § 22757.12 (West 2026) (regulating the assessment and disclosure of catastrophic risk by developers of frontier models).
[102] 49 U.S.C. § 30111(a) (directing the prescription of motor vehicle safety standards). 21 U.S.C. § 393(b)(2) (stating the mission of the Food and Drug Administration). Greenman, 59 Cal. 2d at 62–63 (stating the rule and purpose of strict liability in tort).
[103] Cal. Civ. Code § 1714.46(b) (West 2026) (assigning responsibility for harm in an action against the developer). Professional Engineers, 435 U.S. at 695 (reasoning that safety is among the elements of a bargain that competition favorably affects).