Meta’s $17 Billion Escape Hatch – Music Technology Policy

C.S. Lewis’s The Screwtape Letters is a satirical novel framed as correspondence from Screwtape, a senior demon, to his inexperienced nephew Wormwood. Screwtape advises him how to tempt and manipulate an ordinary human. By presenting morality from Hell’s perspective, Lewis exposes temptation, self-deception, bureaucracy, vanity, and corruption through inversion.

There is something almost C.S. Lewis-like about Meta’s new settlement with the states. Not because Mark Zuckerberg has horns—although I wouldn’t take that to the bank (has anyone seen them in the same room at the same time?) Rather because the genius of The Screwtape Letters was that evil rarely announces itself as evil. Screwtape’s advice is subtler than that. Keep the patient comfortable. Give him reasonable explanations. Encourage small accommodations. Turn moral questions into administrative ones. Above all, make sure nobody notices where the road is actually going.

That is a useful way to read the nominally large settlement Meta just reached with certain state attorneys general over entirely believable claims by the AGs that Facebook and Instagram were deliberately designed to commercialize compulsive and addictive use by users, especially children and teenagers. And the more addicted (or “engaged”) the better, the more profitable for Meta.

The headline number is roughly $17 billion—depending on which part of the settlement and which states are being counted—and even that number deserves an asterisk because billions of that nominal payment are contingent on some bizarre events. The settlement also promises various time limits, nighttime restrictions, parental controls, age assurance and an independent auditor.

It sounds formidable unless you read the actual settlement—and then Screwtape begins to smile.  

The comparison of the Meta case with the Big Tobacco settlement from 1998 is not rhetorical excess. The states themselves are making that analogy. The attorneys general describe the Meta settlement as the largest state consumer-protection settlement since the Big Tobacco settlements of the 1990s. They say their investigation found that Meta designed Instagram with addictive features, knowingly exposed young users to serious mental harms and misled the public about the safety of its products.  Readers will recognize this as another day that ends in “y.”

The resemblance to the tobacco settlement is uncomfortable for Big Tech.  The tobacco litigation attacked the product, its design, its addictive characteristics and the business model built around continued consumption.  That is precisely what made this Meta litigation different from the usual fight over online speech: the states were finally attacking the social-media product itself.  This is an approach that I have been in favor of for at least the last 10 years, so I was quite pleased that these cases were getting somewhere.

This complaint by the District of Columbia AG gives you the idea.

That point is essential to understanding what Meta just settled.  This was fundamentally a product liability case.  A products liability claim generally requires proof that the defendant manufactured, distributed, or sold a product; the product contained a design or manufacturing defect when it left the defendant’s control; the defect rendered the product unreasonably dangerous; the plaintiff suffered injury; and the defect proximately caused that injury.  

The federal proceeding in which all of this occurred bears the caption In re Social Media Adolescent Addiction/Personal Injury Products Liability Litigation.  (Settlement Agreement (“SA”) at 6 (PDF p.16).)  That is what made the litigation different—and potentially much more dangerous for Meta.  The states weren’t simply complaining that Facebook and Instagram contained bad material posted by third parties.  They alleged that Meta itself—like Big Tobacco— designed and deployed products with features that encouraged compulsive use by children. California describes the claims as targeting harmful features on Instagram and Facebook that allegedly “drive compulsive use by children and teens to their mental and physical detriment.”

That distinction changes the legal question from:

“Is Meta responsible for what somebody else posted?”

to:

“Is Meta responsible for the product Meta designed?”

Those are very different cases.  The first immediately implicates the protections social-media companies have asserted since 1996 under the safe harbor in Section 230 (47 U.S.C. § 230).  The second question is the traditional product-liability and defective-design litigation and Section 230 does not apply.

If an automobile manufacturer designs an exploding gas tank, we don’t ordinarily begin by asking who supplied the scenery the driver happened to be looking at when the car exploded.  We ask whether the manufacturer designed a dangerous product.

That is essentially the conceptual move these social-media plaintiffs were making. Who designed the recommendation system? Who selected the engagement objectives? Who decided what behavior the system should optimize? Who designed the notifications? Who created infinite scroll? Who conducted the A/B testing? Who possessed the internal data showing how children responded? Who continued operating those features after learning what they did?

Those aren’t principally questions about a third-party speaker.  They are questions about the manufacturer.  And suddenly the tobacco analogy becomes considerably more precise. The great tobacco cases weren’t principally about whether a particular cigarette advertisement contained protected speech. They were about companies accused of designing, manufacturing and selling an addictive product while knowing things about that product that consumers did not.  The Meta cases threatened to ask the digital-age version of the same question.

For perhaps the first time, one of the world’s largest technology companies faced the prospect of defending the design of the social-media product itself before a jury rather than endlessly debating responsibility for somebody else’s content.  Reuters aptly calls the trial “one of the highest-profile tests yet” of the theory that social-media companies themselves harmed young users.  

And then, before Mark Zuckerberg took the stand, abracadabra, Meta settled.  That should affect how we evaluate every loophole that follows in the list below. Because the question isn’t merely whether the settlement pays money to states and gives parents some useful new controls.  The question is whether the settlement actually fixes the defective product.  Or whether Meta paid billions of dollars to avoid having a jury answer that question.  Spoiler alert:  I think that’s exactly what happened and I’m sure I will not be alone in that view.

That brings us to Anderson v. TikTok, Inc., 116 F.4th 180 (3d Cir. 2024) that we previously covered.  Ten-year-old Nylah Anderson died after attempting the notorious “Blackout Challenge,” which encouraged children to choke themselves until they lost consciousness.  According to the complaint, Nylah didn’t go looking for the challenge. TikTok’s recommendation algorithm selected a Blackout Challenge video for her personalized “For You Page.”

The importance of Anderson isn’t simply its horrifying facts.  It is the court’s treatment of the product.  TikTok had the brass to argue—unsuccessfully—that Section 230 protected it because someone else created the underlying blackout challenge videos. But the Third Circuit concluded that Section 230 didn’t necessarily dispose of claims directed at TikTok’s own recommendation of that content. The recommendation system was not merely somebody else’s speech magically passing through a neutral pipe.

It was something TikTok itself designed and operated.  The algorithm is part of the product.

Now read the Meta settlement.  Its marquee redesign protection is a two-hour daily limit for teen users.  But certain functionality, such as time spent engaging in messaging, doesn’t count toward that limit.  SA § II.B.3(a)(i), at 20 (PDF p.30).

The settlement’s definition of “Meta SMP” expressly excludes “any and all … direct messaging features linked to or included in those platforms, such as Instagram Direct or Messenger.”  (“Meta SMP” shall mean (i) the existing and future web and app versions of Instagram (“Instagram”) and Facebook (“Facebook”), (ii) any New Meta SMP, within 30 days following verification that the relevant product or service meets the definition of New Meta SMP by an independent third party, and (iii) any Meta Spin Off App, excluding, in each case, any and all (A) direct messaging features linked to or included in those platforms, such as Instagram Direct or Messenger and (B) for the avoidance of doubt, (1) any other product or service of Meta or any of its Affiliates as it exists as of the Effective Date or (2) any other product, service or feature of Meta and its Affiliates that does not otherwise meet the definition of a New Meta SMP regardless of its integration with Instagram or Facebook (other than any Meta Spin Off App).”). SA § I.II, at 6 (PDF p.16).

Think about that exclusion in light of Anderson.  Dangerous social-media challenges don’t necessarily remain confined to an algorithmic recommendation feed. Once discovered, they can be sent to friends, forwarded, discussed, imitated and propagated through direct and group messaging. A child can discover dangerous content through an algorithmic feed and then move the interaction into messaging.  Another child can receive dangerous material directly from another user. A group of teenagers can discuss and coordinate a challenge through a group chat.

And under the Meta settlement, the clock apparently stops counting when the teenager moves into messaging.

Indeed, after the daily limit is reached, the agreement expressly contemplates messaging remaining available. Meta may display a neutral screen saying that other parts of the Meta product remain available, although it cannot affirmatively recommend that the teen switch to messaging. SA § II.B.3(a)(ii), at 20–21 (PDF pp.30–31).  And you know that’s exactly the kind of scam Zuckerberg is known for.

Consider the bizarre consequence.  A teenager could exhaust the supposedly protective two-hour daily allowance and nevertheless continue communicating through Meta’s own product.  If a dangerous challenge, self-harm discussion, bullying campaign, sexual solicitation or other harmful activity has migrated into messaging, the settlement’s marquee time restriction may no longer restrict the child’s exposure to it.

That exposes the conceptual weakness of regulating screen time instead of dangerous product pathways.  Anderson shows why those pathways undermine the headline restriction in the settlement. The Blackout Challenge was dangerous because of the conduct it induced. The recommendation system mattered because it placed dangerous material in front of a particular child.  But after dangerous material enters a social network, algorithmic recommendation is only one means by which it can spread.  Messaging is another.  So why doesn’t messaging count toward the two hour cap?

Imagine resolving the tobacco litigation by limiting teenagers to two hours of cigarette smoking per day—but providing that cigarettes smoked while talking with friends don’t count.  That wouldn’t be a meaningful restriction on the dangerous product.  It would be an accounting convention at best. And that may be the first great weakness in the Meta settlement: it measures a teenager’s time in particular Meta interfaces instead of comprehensively addressing the mechanisms through which Meta’s product can expose a child to harm.  It’s treated as a feature, not a bug. Want to bet that’s not the conclusion that the average person would draw without the time to pick into the settlement agreement and who just read the press releases?

Anderson tells us why that distinction can have consequences measured in something considerably more important than minutes.

Messaging isn’t the only exclusion.  The settlement also excludes time spent watching “Longform Content.”  And “Longform Content” includes video or audio lasting at least 22 minutes.  SA § II.B.3(a)(i), at 20 (PDF p.30).

So the headline is:

META AGREES TO TWO-HOUR DAILY LIMIT FOR TEENS.

The fine print is:

Except messaging.  Except qualifying longform content. Except accessing settings. That isn’t necessarily a two-hour limit on Meta use. It is a two-hour limit on certain kinds of Meta use.  There is a difference.  And Phase II doesn’t cure the problem. Its 60-minute-per-platform/120-minute cumulative limit contains the same exclusions for Longform Content, messaging and settings. SA § II.B.3(b)(i), at 21 (PDF p.31).

This loophole is where remembering that we are dealing with a product-liability case becomes especially important.  Teen users must be offered a “Non-Personalized Feed.”  Excellent. Except Meta isn’t required simply to turn off personalization universally.  The personalized recommendation machinery survives.  And my bet is that if it survives, it will be used. That is relevant because the states’ central allegation wasn’t merely that teenagers spent too much time looking at screens.  It was that features of the product itself were designed to drive and amplify compulsive use aka “engagement”.

So if recommendation, personalization and engagement optimization were components of the allegedly defective product, the obvious product-liability question is:  Why are they still there?

The settlement doesn’t dismantle the machine. It installs a timer beside it, and then excludes certain uses from the timer.

Many headline protections are defaults rather than immutable product restrictions.  Parents can authorize less restrictive usage settings and modify certain restrictions.  And my bet is that if anyone can defeat the restrictive usage, it won’t take long for that cheat to become common knowledge. There are obviously legitimate reasons to give parents authority over their children.  But from the standpoint of product safety, this produces a peculiar result.

Imagine resolving a lawsuit alleging that a manufacturer deliberately engineered a product to create compulsive behavior in children by requiring the manufacturer to install a safety device that an “adult” can deactivate. Parental controls themselves are hardly invulnerable. Kids can migrate to secondary accounts, browsers or other devices, or obtain the credentials necessary to alter settings. Meta recognizes at least part of this problem: the agreement requires linked accounts to share the daily limit and requires Meta to use and improve controls designed to verify that the purported supervising adult really is the teen’s parent or guardian. SA §§ II.B.3(a)(v), II.G.8, at 21, 29 (PDF pp.31, 39).  Good luck with that.

Loophole No. 5: Meta Gets to Keep Part of the Headline Money

The “$17 billion settlement” isn’t simply a $17 billion check. The guaranteed amount is substantially lower, with billions more contingent on comparable restrictions applying to competitors. Reuters reports the overall settlement in the range of roughly $17 billion to $18 billion depending on the components counted, while other reporting puts the base multistate amount at approximately $12.1 billion with roughly another $5 billion conditional. This is how the DC attorney general explains it:

In total, Meta, over a ten-year period, will pay at least $12.1 billion to resolve the states’ lawsuits. The total dollar amount is contingent on future settlements with other social media companies. Meta will pay an additional $5 billion—increasingly the total settlement amount to $17.1 billion—and enhance the new safety features it is implementing if and when other major social media companies also agree to adopt these features.

The agreement distinguishes Guaranteed Installment Payments from Contingency Installment Payments. SA § VI.A–D, at 37–38 (PDF pp.47–48).  The contingent money becomes payable only if the defined “Contingent Monetary Payment Trigger” occurs. That trigger is itself tied to “Industry-Wide Adoption” and monetary obligations imposed upon Meta’s “Core Industry Members.” SA § I.U, at 3 (PDF p.13).

But wait, there’s more.  If a settling state fails to achieve the trigger during the ten-year Agreement Term:  “the Contingency Installment Payments shall be permanently forfeited by such Settling State and retained by Meta.” SA § VI.D.3, at 38 (PDF p.48).

So at least part of that enormous headline number is money Meta may never have to pay.  Remember that whenever somebody calls this a “$17 billion settlement.”

Perhaps the strangest provision in this products-liability settlement is something called “Industry-Wide Adoption.”  Some of Meta’s tougher obligations don’t become operative simply because the states alleged that Meta designed a defective product. They instead depend upon Meta’s competitors becoming subject to comparable restrictions, a kind of perverse MFN.

The settlement defines “Industry-Wide Adoption” by reference to the Core Industry Members— TikTok, Snap, and YouTube/Google, and certain qualifying new entrants—becoming subject to specified time-management and age-assurance requirements, including independent auditing. SA § I.DD, at 5 (PDF p.15).

But stop and consider the premise buried inside that structure.  Whose defective product was this case about?  If the states’ case was that Meta designed Facebook and Instagram with particular features that caused or contributed to compulsive use and injuries among children, why should Meta’s obligation to correct those alleged defects depend upon what TikTok, Snap, YouTube or anybody else does?

Imagine a defective-product case against an automobile manufacturer:  We allege that your accelerator is dangerously designed. You must redesign it—but only after Ford, GM, Toyota and Volkswagen agree to comparable accelerator restrictions. Nobody would mistake that for ordinary products-liability relief where the manufacturer whose product allegedly caused the injury would be required to address its own product.

“Industry-Wide Adoption” subtly changes the premise of this litigation. It moves the discussion from remedying Meta’s allegedly defective product to establishing a common regulatory standard for an entire industry.  And that necessarily raises another question: Did all these companies actually design the same product?

Maybe their products share certain characteristics. Perhaps they use similar recommendation systems, engagement metrics, notifications, infinite-scroll interfaces or other mechanisms.  But similarity is something that should be demonstrated, not presumed and certainly not made into a material regulatory deal point.

TikTok’s recommendation architecture isn’t necessarily Instagram’s. Snapchat isn’t Facebook. YouTube isn’t Messenger. Their products may create different risks, employ different engagement mechanisms and require different remedies.  Or do they?

Yet “Industry-Wide Adoption” starts treating these supposedly competing products as so interchangeable that Meta’s material legal duties or at least obligations can depend upon equivalent restrictions being imposed on the others.  That is particularly odd because the states never obtained a verdict establishing that Meta’s own product was defective in the first place, much less a verdict establishing that the products of Meta’s competitors suffer from equivalent defects.

Remember, the settlement stopped the Meta trial (or at least this Meta trial).  Meta admitted no wrongdoing.  And instead of obtaining an adjudication concerning Meta’s particular design choices, the settlement begins constructing an industry-wide product standard.

There is an additional irony.  Meta’s competitors weren’t defendants in the Meta products-liability trial. They didn’t design Facebook or Instagram. They didn’t make Meta’s internal product decisions. They didn’t conduct Meta’s testing. They didn’t possess Meta’s internal knowledge.

Yet their future regulatory treatment can determine Meta‘s future obligations.  Why?  If the defect is Meta’s, fix Meta’s product.  If the problem instead is an industry-wide product architecture, then establish that through legislation, regulation or separate proceedings examining those products.  But don’t quietly transform one company’s products-liability settlement into an industry-wide regulatory compact, a kind of consent decree by another name.

The settlement goes even further. In determining whether competitors’ arrangements qualify for Industry-Wide Adoption, competing products generally cannot exclude products other than messaging or Longform Content unless Meta agrees in writing, subject to the settlement’s competitive-disadvantage standard. SA § I.DD, at 5 (PDF p.15).

So the defendant accused of designing the defective product obtains a contractual role in determining whether the regulatory treatment of competing products is sufficiently equivalent to trigger its own additional obligations.  That is an extraordinary inversion.  The alleged product defect has become an industry standard, and the alleged manufacturer has become one of the referees.

Combine that with the settlement’s MFN provision and the structure becomes even stranger.  Instead of asking the traditional products-liability question—What must Meta do to make Meta’s allegedly defective product safer?—the agreement begins asking: What restrictions can we get everybody in the industry to accept without putting Meta at a competitive disadvantage?

Those are fundamentally different questions. One is product safety to protect consumers. The other is market coordination to protect competitors.  But only one of the competitors was supposed to be in Meta’strial.

Loophole No. 7: Meta Gets Competitive Protection Too

The settlement also contains provisions protecting Meta if states subsequently give qualifying competitors materially more favorable treatment.

Think about that.  The company accused of designing an addictive product for children hasn’t merely negotiated the resolution of claims involving its own alleged conduct.

It has negotiated protection against its competitors receiving a better regulatory deal.  That’s not simply settlement language.  That’s oligopolistic competitive positioning.

This one deserves far more attention.  The settlement attempts to anticipate future social-media competitors.  But its framework excludes certain products centered on artificial-intelligence features, chatbots or interactions with artificial intelligence.

Pause there.

We are settling litigation alleging that one generation of interactive technology used personalization and behavioral engineering to keep children engaged while carving an emerging generation of interactive AI products out of the definition designed to capture future social-media competitors.  For anyone watching the convergence of social platforms, generative AI, synthetic companions, personalization and behavioral modeling, that should be circled in red.

Screwtape has discovered chatbots.

The settlement provides for an independent auditor.  That sounds meaningful, but if you have ever audited Big Tech you’ll know that they treat such audits like litigation and can drag them out, potentially for years.  Compliance auditing going forward is very different from publicly reconstructing the historical conduct that produced the litigation.  

And perhaps the greatest nonmonetary benefit Meta bought isn’t listed anywhere in the settlement consideration:

The trial stopped.

Mark Zuckerberg had been expected to testify.  Then the case settled and the trial ended.  Old CEO tricks.  And Zuckerberg didn’t have to take the witness stand. Contemporary reporting specifically noted that he had been expected to appear before the jury.

That changes how we should think about the price.  Meta didn’t merely buy peace with the states.  It bought the end of an extraordinary public examination of its product.  It stopped additional testimony.  It stopped additional internal documents from becoming trial exhibits.

And it prevented one of the world’s most powerful corporate executives from sitting before a jury and answering questions under oath about what his company knew about the design and effects of its products.  That is particularly significant in a product case.  Because knowledge is the ballgame.

What did Meta know?  When did it know it?  What did its own research show?  Which product-design decisions followed that research?  Which didn’t?  And perhaps most importantly, who made those decisions?

Those are precisely the kinds of questions a product-liability trial can force into the open.  Which brings us back to tobacco.

One of the defining images of the tobacco era shows seven tobacco executives standing before Congress with their right hands raised.

One of the congressmen questioning them was then-Representative-now-Senator Ron Wyden.

Wyden went down the line asking whether the executives believed nicotine was addictive.  One after another, they said “no” which was ridiculous but excellent Washington theater.  Imagine if that happened to social media executives? Wyden understood something important: when an enormously profitable industry stands accused of selling an addictive product while denying or minimizing the dangers, put the people running the companies under oath and ask them what they knew.

That makes today’s Section 230 debate deeply ironic.  Wyden co-authored Section 230 and remains one of its most important congressional defenders, indeed one of Big Tech’s biggest defenders.  Wonder why? 

There is an important legal qualification. Section 230 isn’t absolute immunity for everything a platform does. Anderson itself illustrates why.  But perhaps the lesson from tobacco should have been broader.

When the product itself has been engineered to keep the customer consuming or “engaged”, society should be extraordinarily skeptical of legal rules that transform questions about product design into questions about somebody else’s speech.

Ron Wyden understood the principle perfectly well when the allegedly addictive product delivered nicotine.  The question is why Washington has had so much trouble recognizing it when the allegedly addictive product delivers engagement.

There is another question the settlement cannot answer.  Who really decided that billions of dollars were worth paying to stop this product-liability trial before Zuckerberg testified?  Meta’s corporate structure makes the question unusually interesting. Zuckerberg controls Meta through its dual-class stock structure. His supervoting shares give him majority voting control despite holding a much smaller percentage of the company’s economic equity.  (Just like Daniel Ek at Spotify.)

There is presently no public evidence establishing that Zuckerberg personally ordered this settlement, much less that his supervoting shares were formally invoked to approve it.  We shouldn’t pretend otherwise. But that isn’t the only relevant question.  The governance question is whether Meta’s extraordinary concentration of voting control altered the ordinary corporate checks surrounding an enormous litigation decision that happened to prevent the controlling shareholder, chairman and CEO from testifying under oath.

At an ordinary public company, shareholders ultimately possess the franchise through which directors and management can be held accountable.  At Meta, Zuckerberg controls that franchise.  So when Meta makes an enormous litigation decision whose consequences include preventing Zuckerberg from taking the stand, it is entirely legitimate to ask:

Could anyone at Meta realistically have told Mark Zuckerberg that the shareholders’ interests required Mark Zuckerberg to testify?

That question deserves an answer.

The settlement contains a broad governmental release. It resolves an enormous body of state claims and brings years of investigation and litigation against Meta to an end.  Meanwhile, private litigation and other social-media cases continue.  That’s another reason not to mistake settlement for adjudication.  There was no verdict resolving whether Meta committed the conduct alleged by the states.  There was no completed product-liability trial.  There was no Zuckerberg testimony.  There was no completed public trial record.

Meta denies wrongdoing.  And the states took the money and negotiated a kind of consent decree instead.

None of this means the settlement is worthless.  Some provisions are substantial. Age assurance may improve. Overnight access may decline. Notifications will be restricted. Certain appearance filters and visible reaction counts will disappear for teen users. Parents will have better tools. Those are real accomplishments. But that is precisely why this agreement deserves to be read rather than merely celebrated.  Because the central question raised by the litigation wasn’t simply:

How long should children be allowed to use Instagram?

It was much more fundamental:

Did Meta design a dangerous product for children?

And if the product itself was the problem:

What happens when the product is the addiction machine?

The settlement largely answers different questions: How many hours may the machine operate?  Which portions of the machine count toward those hours?  Does messaging count?  Does longform video count?  Can parents loosen the restrictions?  Does the personalized recommendation system survive?  Must competitors adopt similar rules?  Do AI interactions count as social media?  How much does Meta pay if everybody else eventually joins the deal?

Those are important questions.  But they aren’t the question a product-liability jury was about to answer. 

And perhaps most importantly:

What did Meta know about the machine, who knew about it, and why did the trial end just before Mark Zuckerberg had to answer questions about it under oath?

C.S. Lewis’s Screwtape understood this kind of moral substitution.  The safest road is the gradual one.  No sudden turns.  No milestones.  No signposts.

And, if possible, no controlling CEO on the witness stand explaining exactly what his company knew about the product it built.

Screwtape is smiling.