What K.G.M. v. Meta Platforms Means for Liability and Free Speech

In K.G.M. v. Meta Platforms, Inc., a 20-year-old woman alleged that she became addicted to Instagram and YouTube after prolonged use starting from childhood. She argued that infinite scrolling, autoplay, and algorithmic recommendations are designed to keep young users engaged and that they contribute to serious mental health injuries. Meta and Google were found negligent in the design or operation of Instagram and YouTube, with the jury concluding that both companies failed to provide adequate warnings and awarding $6 million in compensatory and punitive damages.

But even if the verdict is ultimately overturned, the theory behind it matters because K.G.M. may provide future plaintiffs a legal path around such cases’ traditional focus on online content by targeting a platform’s design and business decisions. Conversely, if courts allow plaintiffs to characterize decisions about recommending, ranking, or presenting speech as mere “design,” then the same theory could become an increasingly broad way to bypass First Amendment protections for online expression— including for generative artificial intelligence (GenAI).

The Case for Platform Design Liability

For years, internet liability disputes largely revolved around content: who created a harmful post, who displayed it, and whether a platform could be treated as the publisher of someone else’s speech. Section 230 both prevents an interactive computer service from being liable for third-party content and grants them a safe harbor for content moderation decisions.

Rather than blaming one post or video for causing harm or distress, the plaintiffs in K.G.M. argued that the platforms created an unsafe environment for their users by design. They claimed that the system was designed to encourage prolonged and compulsive use due to features like infinite scrolling without natural stopping points, autoplay without a new user decision, and recommendation systems continuously selecting material intended to hold the user’s attention.

In a November 2025 pretrial ruling, Los Angeles Superior Court Judge Carolyn Kuhl decided to allow a jury to conclude whether certain design features could cause harm independent of the specific content viewed. The court explained that companies may still face liability for harms arising from the design of their platforms simply because those design features also affect how information is communicated. In other words, the fact that a challenged feature has a communicative function does not necessarily shield the platform from liability for harms allegedly caused by the feature’s design. The jury ultimately agreed, finding that the platforms failed to exercise reasonable care in designing features that a reasonable person could foresee might be harmful to young users.

The Limits of Platform Design Liability

K.G.M. echoes Lemmon v. Snap, Inc., a Ninth Circuit case that allowed a product liability claim to proceed against Snapchat. In that case, the plaintiffs alleged that Snapchat’s Speed Filter, which overlaid a user’s travel speed on any photo or video—combined with a reward system that encouraged users to share high-speed content—foreseeably encouraged dangerous driving and crashes. The Ninth Circuit specifically concluded that the duty to design a reasonably safe product was independent of Snap’s role in monitoring or publishing third-party content.

Lemmon provides an important limiting principle for cases like K.G.M. A platform should not escape ordinary negligence law simply because a dangerous feature happens to exist within a communications service. But courts should require plaintiffs to identify an independently harmful design decision rather than allowing “design” more broadly to become another vehicle for challenging the dissemination of lawful speech.

The problem arises when the alleged defect cannot be separated from the information being communicated. Recommendation systems and ranking mechanisms do not merely affect how long users remain on a service; they also determine which information reaches an audience. If liability arises whenever a recommendation system exposes someone to harmful or objectionable material, companies have an incentive to recommend less content, remove more borderline material, and limit access to controversial but lawful speech.

Negligent Design Requires More Than Harm

While K.G.M. did not declare Instagram and YouTube defective products in every circumstance, the verdict reflects the court’s belief that companies can be liable for failing to use reasonable care during design. Courts should keep that theory narrow—claiming a platform is “addictive” should not be enough to win a trial.

To meet the legal standard for negligent product design, plaintiffs should have to prove that: 1) any reasonable person could foresee that the design could be harmful to users; and 2) that the flawed design directly caused harm. Plaintiffs should have to identify a specific design choice, a foreseeable risk, a reasonable precaution, and a causal connection between the alleged failure and the injury before making a claim.

Causation should be a major limit. Mental health injuries, for instance, rarely have one source. The sheer fact that a troubling feature exists does not prove that it caused a particular injury.

Applying Platform Design Liability to AI

The same framework should apply to GenAI. Although developers cannot escape all responsibility for safety merely because their choices were implemented through code, most AI development should be considered editorial decision-making, which is covered by the First Amendment. The question is whether an alleged harm arose from a design decision that independently created a foreseeable risk or whether a plaintiff’s claim merely objects to what an AI system communicated.

A developer marketing a chatbot as a companion or source of emotional support may have a harder time claiming it assumed no responsibility for foreseeable reliance; however, K.G.M. demonstrates the risk of expanding that theory too far. Almost everything an AI system does ultimately reflects some design decision—from its training, system instructions, and safety mechanisms to its memory, interface, and rules governing how it responds. If the fact that an output resulted from “design” is enough to avoid traditional protections for speech, the distinction between content and conduct could lose all meaning.

The practical consequence would be pressure on AI developers to make models more restrictive. Faced with uncertain liability, companies may limit responses involving controversial subjects, reduce access to sensitive but lawful information, or disable useful features altogether. Courts must therefore make a meaningful distinction between a defective mechanism that independently creates a foreseeable risk and a claim that ultimately objects to AI-generated speech.

User Conduct Must Factor Into Liability

Comparative fault will remain a strong defense against broader claims. Companies should not be expected to prevent every determined user from circumventing a model’s protections. If a user repeatedly and deliberately defeats meaningful safeguards, then that conduct should matter when courts assess responsibility. The relevant question is not whether safeguards can ever be bypassed, but whether the company took reasonable precautions against foreseeable risks and whether the alleged harm resulted from the failure of those precautions.

Drawing the Line Between Design and Speech

There should always be edge cases in which a truly malevolent or grossly negligent platform might face liability. But courts should set the bar high to avoid turning “product design” litigation into de facto speech regulation. Although that tradeoff ultimately places more burden on individual users to regulate their own interactions with these platforms, it is a better alternative to judges dictating the shape of emerging technologies.

Liability is similarly plausible when a claim challenges independently wrongful conduct, such as knowingly ineffective age protections, deceptive safety claims, or failure to activate promised crisis tools. However, claims deserve greater scrutiny when they would require courts to dictate what lawful information a platform may communicate, how it must rank competing ideas, or which viewpoints an AI system must favor.

That distinction matters because Section 230 was designed to prevent platforms from being treated as the publisher or speaker of third-party information. If a plaintiff can avoid that protection simply by describing recommendations, ranking, or other publishing functions as “defective design,” then the exception risks swallowing the rule.

The effects would extend beyond the companies defending these lawsuits. Large platforms may be able to absorb years of litigation, but smaller platforms and emerging AI developers have fewer resources to do so. Faced with uncertain liability, the rational response may be to remove more material, recommend less information, and place tighter restrictions on what users can access. Liability intended to address harmful design could therefore indirectly restrict lawful online speech.

Hopefully, K.G.M. does not mark the end of Section 230 or let every digital service be sued as defective products. The case proves that the distinction between publishing content and designing the systems around it is no longer theoretical. The law can and should hold companies responsible when their independently negligent choices create foreseeable harm; however, courts should not allow every dispute over online speech to become a product liability issue.

The challenge is preserving both principles at once: A company should not be able to hide negligent product design behind Section 230, and a plaintiff should not be able to evade Section 230 merely by calling a publishing decision a design defect. Without that boundary, “design defect” risks becoming a side door for regulating online speech.