A Close Read of Big TikTok and Meta Settlements Raises Important Questions

A collage of circuit boards interwoven with aerial photographs of water treatment facilities, agricultural lands, industrial sites and mining machinery.

Brazil's AI Development Strategy Needs an Edge Layer

Anthropic's logo is displayed on a smartphone with a stock market chart in the background on October 27, 2026, as Anthropic filed a Form S-1 with the Securities and Exchange Commission (SEC) for a planned initial public offering in the United States. (Photo by Samuel Boivin/NurPhoto via AP)

Could the United States Seize Anthropic at a Discount?

Journalists test AI software used to manipulate images at Phnom Penh Municipal Police headquarters following an operation against scam centers in Phnom Penh, Cambodia, Tuesday, Sept. 22, 2026. (AP Photo/Heng Sinith)

Cambodia’s Scam Crackdown Has Failed to Stop the Industry’s Global Spread

Two women sitting on a bench smiling with a US flag and an 'I voted' button

Photo by Pocstock, an inclusive, BIPOC-owned media and stock photo company that features real, human-captured photography.

Tech Bros Are White, Progressives Afraid, But Black Women Public Interest Tech Leaders Are Brave

The TikTok logo is displayed through a magnifying glass in this photo illustration in Ontario, Canada, on February 5, 2026. (Photo by Thomas Fuller/NurPhoto via AP)

Can the EU Make TikTok Take Responsibility for Its Rabbit Holes?

In the foreground, multiple people are surrounded by computer screens filled with data visualizations, charts, and technical information. A woman seated in an armchair appears deep in thought, surrounded by data-filled monitors. Beside her, a man leans over, using a tablet to assist with their inspection of a plant or tree. In the centre, a figure holds a large frame or screen displaying anatomical illustrations, representing the use of AI to analyse medical imagery. To the left, another person is intently observing a computer screen, while a second figure nearby is deeply immersed in analysing data. A woman dominates the right side of the composition, gazing upwards as if in contemplation or envisioning something beyond the immediate scene. The background features more people, including a family holding hands, and other abstract representations of data.

AI is Everywhere by Ariyana Ahmad & The Bigger Picture / 

In the AI Era, Language is Now an Attack Surface

Close up photo of a pile of US 25-cent coins

10 Things States Should Do with the Meta Settlement Money

The sun rises over the US Capitol and the Washington Monument on the National Mall the day after election day on November 04, 2026 in Washington, DC. (Photo by Kevin Dietsch/Getty Images)

A New Agenda for Tech Regulation Beyond AI

People attend a watch party for Georgia Gov.-elect Keisha Lance Bottoms at the Atlanta Marriott Marquis on November 3, 2026 in Atlanta, Georgia. Bottoms defeated Republican gubernatorial candidate Rick Jackson to become the nation's first Black woman governor. (Photo by Megan Varner/Getty Images)

US Midterms Results Present 'Real Moment for Change'

A woman wears a mask with a Wi-Fi symbol as a protest against internet censorship in Iran. Several hundred people demonstrated in Frankfurt, Germany on Rathenauplatz on May 16, 2026, demanding an end to the Islamic regime in Iran on placards and in chants. (Photo by Boris Roessler/picture-alliance/dpa/AP Images)

Freedom House Charts 'Seismic Shifts' on the Global Internet

Image of map and computer language over binary codes on server racks in server room. 

Global Digital Policy Roundup: October 2026

A protester holds a Fight Data Centers signs Monday, Oct. 12, 2026, in Mason, Mich. (AP Photo/Paul Sancya)

Assessing the Impact of the Midterm Elections on US Tech Policy

Protesters fill a city street holding banners and signs. A large pink banner in the foreground reads “Stop Dirty Data Centres.” A teal banner reads “Save Brick Lane.” A handwritten sign reads “Stop Social — Build Homes Not Data Centres.” Demonstrators carry union flags and drums; a man with a megaphone addresses the crowd at right. Scaffolding covers a brick building in the background.

Protesters march outside the South by Southwest festival in London on June 4, 2026, opposing a planned data center development at Brick Lane in east London. The demonstration was part of a series of actions organized by the “Stop Dirty Data Centres” campaign, which is holding Europe-wide days of action this weekend involving more than 100 groups across 15 countries. (Global Action Plan)

Europe's AI Ambitions Meet the Data Center Backlash

View at the Google DeepMind logo at the office building in London. (AP Photo/Alastair Grant)

The Google Whistleblower Taking on Big Tech Over Its Role in Gaza

A sign supporting the banning of data centers is seen outside of a home near the Meta Prometheus Campus, an AI data center and supercluster project, is seen on November 1, 2026 in New Albany, Ohio. (Photo by Jon Cherry/Getty Images)

Data Centers Were on the Ballot Across the US, and They Lost. What Comes Next?

Elon Musk listens as President Donald Trump speaks during an event at the US Institute of Peace, Oct. 8, 2026, in Washington. (AP Photo/Julia Demaree Nikhinson, File)

Pull the Plug on Technofascism Before It Reboots

Exterior view of the US headquarters of TikTok in Culver City, California, in 2022. Shutterstock

Exterior view of the US headquarters of TikTok in Culver City, California, in 2022. 

Social Media Addiction Cases Are Settling Before Trial. What Does It Mean?

Online Radicalization Concept with Digital Propaganda, Social Media Manipulation, Extremist Influence, Algorithmic Amplification, Hate Speech, Misinformation, Polarization and Online Safety Risks

Hate Speech Still Pays on YouTube. Regulators Should Follow the Money

Anti-data center demonstrators protest at the Pennsylvania Capitol, Tuesday, Oct. 6, 2026, in Harrisburg, Pa. (AP Photo/Marc Levy)

Voters Handed Congress a Mandate to Confront Corporate Power and Technofascism

 its child safety litigation with US state attorneys general, with great fanfare. It committed to make a series of product changes, to pay the states at least $12 billion, and to pay them an additional $5 billion if they convinced Meta’s competitors to join in and make similar commitments. That mechanism — a corporation paying state officials a bounty to impose specific rules on its competitors — is not one I’ve found any precedent for.

A month later, TikTok made the next move, 

 litigation with the Alabama attorney general. It makes similar-looking product concessions, sometimes using language identical to the Meta agreement. Its payments take tech platform cooptation of government power to a new level. Alabama gets 

 up front, but will more than double its payout — collecting another 

 from TikTok — if enough other attorneys general settle with the platform.

TikTok’s settlement has drawn far less attention. It was filed at 4:30 p.m. on a Friday at the end of summer. The official 

 could not be searched, annotated, or even easily converted to other formats. (A cleaned-up, annotated copy is 

.) The settlement’s product commitments are also pretty weak compared to Meta’s. This piece will explain why, and compare TikTok’s settlement in broad strokes with Meta’s. More details about Meta’s settlement from Electronic Frontier Foundation senior counsel David Greene are 

This is a long-ish read, but it has clear section headings to help busy readers skip to the parts they need. I did my own reading and reached my conclusions for this the old-fashioned way, but I used Claude quite a bit to check my work. To the extent it provided useful additional answers, I put them 

My goal is to increase the number of people who understand the settlements, because they are building blocks of our future online. These two documents may be the subjects of major First Amendment litigation. They may be emulated around the world and define the invisible, state-backed rules shaping information access for a generation of young people in an era of global 

. Clear-eyed understanding of the agreements’ intricacies will matter.

These settlements reflect two larger issues. First, Congressional inaction has spurred fragmented efforts by individual states to address child safety issues through new laws or lawsuits. Second, the First Amendment bars many of the laws proposed by state legislators, and many of the legal claims advanced by attorneys general. Private settlements with platforms offer an apparent work-around — a way for governments to achieve outcomes that, if mandated by law, would be unconstitutional.

These backroom deals between state officials and private companies provide no opportunity for meaningful public oversight or challenge by affected internet users — including children. As I will discuss in a forthcoming paper, the result is a set of platform commitments that may or may not make kids safer, but come with major downsides for free expression, privacy, and competition. We can’t begin to assess those tradeoffs without being clear about what the settlements actually say.

Both settlements are long and detailed. Copies with decent OCR and section headings in the left navigation bar are 

 for TikTok. AI-generated answers to some in-the-weeds questions are 

 tracking some differences in these and possibly future cases and settlements is here.

Meta’s settlement resolved child safety claims with 51 attorneys general (representing 47 states, plus DC, the Northern Mariana Islands, American Samoa, and Puerto Rico). The per-state payout varies greatly, but on average the 51 attorneys general get about $235 million now, and an eventual potential total of $333 million. The three states that did not join the settlement are 

, which reached a separate settlement for $1 billion; 

, which has won a series of cases against Meta including a staggering 

, which is still litigating and accused the others of settling for “peanuts[.]” Most of the Meta settling attorneys general, including Alabama, also settled unrelated claims dating to the Cambridge Analytica privacy brouhaha, getting $459 million to split among them.

Meta commits to make a number of product changes (discussed below) and submit to extensive audits. It also agrees to additional “Phase II” product changes that will further restrict teens’ time on the platform if certain contingencies are met. Meta’s potential future commitments can be triggered on a state-by-state basis as follows:

, preventing teens from using most features for nine hours overnight instead of six hours, if specified 

 (1) adopt those same nighttime limits, (2) adopt age assurance measures at least as stringent as Meta’s under the settlement, and (3) undergo independent audits for these commitments. (Meta’s own audit commitments are extensively spelled out, but competitors’ exact obligations are less clear.)

, shifting teens from a 2 hour cap on cumulative use of Facebook and Instagram in each 24-hour period to separate 1 hour caps on each platform if competitors (1) adopt the same daily-use limits along with (2) age assurance and (3) auditing.

 in contingent payments if competitors make these product changes and 

 make their own payments to the settling states. Each competitor’s payment to a settling state must be at least as much as Meta’s contingent payment. For example, since Meta would pay Alabama about $50 million in this scenario, each competitor would also need to give the state $50 million. Platforms with profits that do not exceed $10 billion/year are exempt from this.

The listed competitors (or “Core Industry Members”) that must sign on for this to happen are TikTok, Snap, YouTube, and potential 

 market entrants over a certain size. Their adoption of the nighttime and daily-use limits establishes what Meta calls “

. Meta’s initial commitments for nighttime access limits and daily-use caps last 

 years, but the window of opportunity for competitors to join and trigger Phase II’s heightened version of those obligations lasts for 

 specifies that the attorneys general can get more money from Meta if it agrees to pay more per capita in a pre-trial settlement with another state. (It’s too late for a settlement in New Mexico’s current cases to trigger this new round of payouts, but what about 

? Negotiations there could get interesting. (Claude’s take is 

.) Settling states also get parity if Meta promises another state certain more restrictive screen time limits.

On the flip side, Meta gets benefits if a settling state (including Alabama) imposes lower obligations on a competitor (including TikTok). In principle, Meta can ask for — and Alabama must agree to accept — the more lenient obligations TikTok assumed for auditing, age assurance, night mode, daily caps, school mode, productive pauses, and identification of secondary accounts. Changing those things just for Alabama may not make sense as a product matter for Meta. But TikTok’s obligations are a lot more lax, especially for auditing. This is a weird hole for Alabama to drive through the Meta settlement.

TikTok, too, agrees to an immediate payment and product changes. As discussed below, most of these product changes look similar to Meta’s, but some are meaningfully weaker, and the lack of auditing removes a key mechanism for accountability and enforcement. It is also unclear what obligations, if any, TikTok assumes for users under the age of 13 aside from checking their age.

TikTok’s agreement contemplates future contingent, heightened obligations, much like Meta’s Phase II. The triggers for (1) product changes, (2) additional payment to Alabama, and (3) auditing are separate.

Like Meta, TikTok will make additional product changes triggered by competitors’ commitments. In TikTok’s version of Phase II, its competitors must commit or be legally required to do the following, at which point TikTok will do the same.

 certain “numerical representations or quantification of reactions on content” by default. (This is the most visible difference from Meta’s Phase II. As discussed below, this differs from Meta’s commitment to hide likes in both timing — by happening only at Phase II — and in important implementation details.)

 access for nine hours. (For TikTok, like Meta, this extends an initial six hour commitment.)

 caps analogous to Meta’s Phase II rules. (For TikTok, this could be no change at all. The agreement states that if Meta gets two hours of teen use across both its platforms in Phase II, TikTok still gets two hours on its single platform.)

 for kids 13-17 under TikTok’s negotiated standards. (This is not a big substantive change for TikTok, Industry-Wide Adoption of its standards may cause its commitment to last longer.)

Somewhat oddly, although the settlement sets rules for other companies to adhere to, it also 

 to be “not intended for use by any third party for any purpose, including submission to any court for any purpose[.]”

TikTok will pay Alabama up to $183.8 million — roughly 1.5x its initial payment — if enough other attorneys general settle with TikTok. These payments are split into four potential stages, each lasting up to two years. In each stage, ten more attorneys general must sign “qualifying agreements” with TikTok in order to trigger that stage’s payout and start the two-year countdown for the next one. Failing to recruit enough attorneys general in any stage causes any future chance of payout to “permanently lapse, terminate, and be forfeited[.]” Alabama can also get additional future “

” payments if other attorneys general who took part in Meta’s settlement negotiate high enough payments from TikTok. If those officials collectively get over $5.1 billion, any overage triggers a payout to Alabama of 1.55%, a figure representing Alabama's share of the national population.

The settlement provides that no more than 25% of Alabama’s payments can go to litigation and investigation costs and attorneys’ fees. (Of Alabama’s $116.2 million initial payment, $16.2 million is earmarked for the 

 who handled the case and litigation costs for the attorney general’s office. Here is some old but relevant New York Times reporting on attorneys general offices’ relationships with private 

Major platforms have recently experienced a surge of audit requirements, including close annual scrutiny of product design and content moderation under EU law. Meta’s settlement agreement commits it to immediately take on yet another audit and lays out detailed auditing provisions. TikTok’s does not. In its initial phase, TikTok will at most do things like “certify” its actions or “provide 

 results to Alabama. This is a major difference from the Meta settlement in both burden and accountability.

 if competitors make the product changes described above and at least 40 attorneys general or those representing 60% of the US population settle with TikTok on terms “substantively equivalent” to Alabama’s daily use cap, nighttime limits, hiding likes, and age assurance for 13-17-year olds. One set of audit 

 will cover detection of users under 13, another will cover 

 (longer night time limits, daily caps, and suppressing likes/reactions). Details of these audits remain to be 

The TikTok settlement states that — other than any ongoing contingent payments for more settlements with attorneys general — it 

 after five years, with some already-existing product commitments staying in effect for 

 years. But some obligations could be in effect as much as ten years in the future. If competitors join in TikTok’s commitments, the second phase product obligations can stay in effect for up to 

, though not past August 26, 2036. (Notably the anniversary of Meta’s settlement — not TikTok’s.) Those obligations 

 if any competitor falls out of compliance. Meta, by contrast, contemplates its Phase II obligations 

 in and out of existence with changes in competitors’ compliance.

If Alabama reaches a settlement with one of TikTok’s listed competitors that is more 

 than this one, TikTok “shall” have an opportunity to accept the same terms (though Alabama only “may” have to offer them).

Has TikTok met the requirements of the Meta settlement?

The commitments that TikTok makes to Alabama are weaker than the commitments required to trigger new consequences for Meta under its settlement. TikTok does not commit to any immediate audits, for starters. And its product changes in at least two relevant categories — nighttime access and daily caps — are probably not strong enough. Its age assurance seems likely not to measure up either. As discussed below, its nighttime rules and daily caps have some key exceptions, including letting kids spend more time on TikTok by searching for content.

It is therefore unclear what Alabama means when, in its Settlement, it “represents that [the settlement] satisfies the conditions applicable to the State under the Meta Consent Judgment.” Perhaps it just means that TikTok’s initial payment to Alabama of $116 million satisfies Meta’s monetary requirement, which I think is correct. (Sufficiently profitable competitors must pay each state the same amount as Meta’s contingent payment, which for Alabama is 

Or perhaps TikTok and Alabama are taking the position that TikTok’s much weaker commitments are actually adequate to trigger Meta’s second round of obligations. That is clearly incorrect, unless Meta itself opts to 

 some settlement requirements. But TikTok has lived with a lot of 

 lately. This one has understandable appeal to any potentially settling platform — and perhaps even to Meta — since it turns on a legal interpretation of Meta’s settlement terms that is extremely lax and platform-favorable.

The TikTok settlement requires the platform to take new measures to protect children. But the potential future legal 

 that Alabama gives up are not just about children. They appear to cover harms to adults, the elderly, or any other kind of person. The state’s 

 cover most liability theories I can imagine about platform design, content moderation, or consumer deception — and very arguably also about privacy. I won’t list all the agreement’s synonyms and variant wording, but Alabama releases any “past, present or future cause of action… of any nature whatsoever” for TikTok’s 

, omissions, negligence, or other shortcomings “relating in any way to the design, development, marketing, or advertising of the TikTok platform or any of the TikTok platform's features, or to any statements about the TikTok platform or any of its features[.]”

It does not release future claims based on new actions taken 

 the settlement date — though the AG might be hard-pressed to argue that ongoing behavior it effectively signed off on in the settlement violates the law. And of course, the settlement binds only the State of Alabama and its subdivisions, officers, etc. That may mean it binds municipalities or school districts, but I do not think it would cover tribal claims and it certainly does not restrict civil claims by private claimants.

It is difficult to compare the released claims in the TikTok settlement to those in the Meta settlement, because Meta’s are fairly open to interpretation. They include any claims that “could have been” made based on the facts alleged in a long 

 of legal filings. I assume that Meta would argue, if pressed, that its settlement also bars attorneys general from bringing most claims about adults. But that would be an argument, while Alabama’s broad releases in the TikTok case are very clear.

There are broad similarities — and many drafting similarities and identical language — in the product commitments made by both platforms. But there are also important differences that make TikTok’s commitments generally weaker, an issue that is compounded by the lack of audits.

Some questions to ask about these commitments, which I won’t explore much here, are

How do these commitments square up against the theories of harm or liability that attorneys general raised initially?

For all the fuss about “Phase II” commitments, are the changes in night mode and daily use that important — to teens, parents, or platforms?

How might commitments have been calculated based on their effect on competitors, as opposed to Meta or TikTok themselves? Anything to do with messaging seems particularly relevant for Snap, and anything to do with Longform video seems particularly relevant for YouTube, for starters.

How should policymakers and others be thinking about the different harms and benefits of consuming content that arrived via a ranked feed, versus a search result, versus a message?

What counts as “messaging” for purposes of the settlements? 

Commitments that are significantly different

Meta has never formally allowed users under 13 (“U13” in settlement parlance) on the platform, and its settlement establishes mechanisms to enforce that existing rule. TikTok does have an “

” version of its service, with limitations including a 60-minute daily usage cap. But the settlement has almost nothing to say about those users or that service — an oddity given that Alabama’s 

 focused in part on its alleged inadequacies. The Under 13 Experience does seem to be formally covered as part of 

, but the settlement’s actual product commitments, like overnight use restrictions or parental controls, are for “

” — defined as 13-17-year olds. No equivalent rules are described for younger users. In other words, under the settlement, TikTok promises protections for fifteen-year-olds that it does not promise for twelve-year olds. That’s a weird outcome.

 methods for U13s. But being identified as U13 leads to “plac[ing] those TikTok accounts into TikTok’s U13 enforcement process” — presumably meaning ensuring that they only see the app’s Under 13 version. The data minimization rules reinforce this, 

 that TikTok may retain data about kids under 13 for purposes of “providing the U13 experience[.]” 

 — both seemingly copied from Meta’s settlement — do refer to efforts to “remove” U13 users. But unless this implicitly means that TikTok is shutting down its Under 13 service, these would seem to just be about removing them from the version available to teens or adults.

Time limits: ‘night mode’ and ‘daily use’

Both of TikTok’s time limit changes will allow kids to use the platform more than Meta’s equivalent commitments. They diverge in a way Meta explicitly anticipated: Its agreement 

 that a competitor may generally not trigger Phase II obligations for Meta if the other platform’s time limits “exclude any features, surfaces, or portions other than messaging or Longform Content[.]” TikTok does exactly that by excluding “Search” and redefining “Longform Content.”

For overnight access, TikTok matches many other aspects of Meta’s commitment. It covers identical blocks of time, expanding from six hours (12 a.m. to 6 a.m.) initially to a potential nine hours (10 p.m. to 7 a.m.) if competitors do the same. Like Meta, TikTok allows overnight access to messaging and settings. That’s a potentially significant exception on 

 platforms, given how much time teens spend sharing videos and other content 

But TikTok adds an additional exception: kids will 

 to use “Search” features at night. It 

 “Search” very broadly, to include access to, viewing of, or interaction with “content surfaced” through searches or “related search recommendations, suggestions, notifications, or similar features[.]” The settlement does separately 

 that TikTok will try to prevent teens from using Search to “access broader TikTok features[,]” but this still effectively looks like a giant loophole. Kids can watch videos at night as long as they arrive via messages or search. Effectively, the thing being blocked at night is not screentime but feed functionality.

 has even more exceptions. Here again, it joins Meta in not counting time spent messaging and accessing settings. But it also doesn’t count “Search” toward the daily usage limit. For daily caps enforcement, TikTok also doesn’t have to try to block Search-based workarounds until the settlement’s second phase. Additionally, while both Meta and TikTok exclude “Longform” content from this cap, for Meta only videos of 

 or longer count as “Longform.” For TikTok, any video 

 or longer is “Longform” and does not count towards the cap.

Meta commits to unconditionally prevent teens “from seeing numbers of likes or reactions on Meta SMPs.” I initially thought this was a pretty meaningless concession. The Wall Street Journal previously 

 that Meta’s internal data showed the change did not improve teen wellbeing — but that Meta nonetheless launched an option to hide likes in 2021 as a bid for better PR. But whether or not this change helps kids, maybe turning it on by default meaningfully hurts platforms. TikTok, at least, seems to see that change as a pain that should not be borne unless the whole industry agrees to do it. TikTok commits to hide reactions only if competitors do the same.

Significantly, TikTok also will hide reactions only in the Friends Feed, or on “content posted by the Teen User from that is displayed on the Teen User’s profile page” (sic). That leaves reactions visible in the app’s main “For You” Feed.

Both platforms will offer algorithmically ranked feeds to teens by default, but provide the option of switching to “non-personalized” home feeds. This may mean very different things for different platforms. The basic user interfaces on Facebook and Instagram currently offer users a single primary feed, so making that non-personalized is a big deal. TikTok lets users switch easily between different feeds, and commits to a non-personalized option on only one — the “home feed[,]” perhaps meaning the For You feed. If that gets boring, personalized options on TikTok are one click away.

The definition of “non-personalized” is also very different in each settlement. For Facebook or Instagram, a “non-personalized” feed appears to 

 a chronologically ordered feed of content from friends or followed accounts. Meta maintains some vaguely worded leeway to intervene for things like “platform integrity[.]”

, a “non-personalized” feed can be chronological — which would probably be a pretty random and unappealing user experience, given how the platform works. Or it can be a feed with more in common with broadcast TV, offering content predicted to appeal to a general audience. TikTok’s wording about ranking based on user data is also confusing: it says the feeds will be ranked using “non-user-specific criteria” and also will not “rely 

 on information associated with a specific user” (emphasis added). It also defines using “social graph signals” as a form of personalization. So the very thing Meta calls non-personalized — a feed of content from friends or followed accounts — is seemingly personalized under TikTok’s definition.

Uncertainty about rules for various algorithmically ranked features seems likely to be endemic when applying either settlement. Both include confusing references to “recommendations,” for example. Is content in a ranked feed a “recommendation”? Stepping beyond Meta and TikTok, would YouTube’s video recommendations be a “feed”? (Claude had some 

 on the settlements’ use of these terms.)

These provisions are not identical, particularly as to kids under 13, though the basic rules look much the same to my inexpert eye. The lack of audits or third-party testing or certification, however, creates a major difference in practice. TikTok also lacks Meta’s rules about using age signals from Apple or Google or setting strict thresholds for vendor tools. This part of the settlement is substantial and deserves to be critically unpacked by someone more specialized in age assurance technology. (Claude’s take is 

Both platforms pledge to protect teens from harmful content, often using identical wording about categories like bullying, nudity, or graphic violence. But both settlements also nominally leave platforms to define those terms for themselves. So perhaps the biggest difference the agreements create is that Meta will be subject to 51 attorneys general asserting their own — potentially politicized — definition of things like “bullying.” TikTok, for now, will be answerable only to Alabama. (This old 

 of mine lists lots of ways that officials who are “just making platforms enforce their own rules” can effectively advance their own preferences about user speech.)

The settlements diverge more in the mechanics of content moderation. As usual, Meta makes stronger commitments — here, on things like notice and action processes, limiting teens’ exposure to third party accounts, and aggregate content exposure (as summarized by Claude 

Both parties also say their content policies are “inspired by movie ratings for ages 13+” — which is funny, given that the MPA sent Meta a 

 once already for claiming to apply PG-13 ratings.

Both settlements allow teens to override new default settings only with parents’ permission. This requires that parents have or create platform accounts, and pair them with teens’ — at which point, the parents get both new controls and new information about things like the people with whom the teen exchanges messages. The details vary. As a parent, I think I like TikTok’s better — it seems to offer more flexibility to do things like set specific time frames for access to the platform. As someone worried about the settlements’ impact on kids with abusive caretakers, though, both versions make me queasy. Meta addresses more operational details and potential avenues for abuse or workaround (as Claude explains 

Both platforms commit to nudge teen users to take breaks at intervals and to impose “productive pauses” at minutes 60 and 90. Meta’s 

 has more mechanisms for defining and refining how pauses will work.

Both platforms agree to prevent teens from using facial filters that show changes only achievable by cosmetic surgery. This requires some rather hard-to-draft exceptions for things like adding cat ears. 

 also bans filters that mimic “extreme makeup” techniques, and gives attorneys general the difficult task of providing Meta with “illustrative examples and guidance[.]”

Keeping a bunch of existing features and educational resources

 of TikTok’s commitments are explicitly identified as things it does already. Meta surely relies on existing practices too, but its drafting makes that less conspicuous.

The data minimization provisions in the two settlements look similar to me, and no other major privacy differences jumped out at me except teens’ slightly greater privacy from parental supervision on TikTok. (Claude identified more differences 

 allowing TikTok to retain data about kids under 13 for purposes of “providing the U13 experience” on the platform.)

Have you read this far and you still want more? Okay, here is a little more, then we really must stop.

TikTok’s obligations under the settlement apply to “the operation of the TikTok platform in the State of Alabama.” I take that to mean that TikTok can geotarget a version of its service that respects the settlement to devices in Alabama. Meta makes more references to “users in the Settling States” — which could still mean geotargeting, but arguably suggests that the same rules apply on a per-account basis when a known teen user from a settling state travels. That may just be an easier way for either platform to handle compliance, anyway.

Given TikTok’s unusual history, the legal entities that created liability may not be the same ones that negotiated or will comply with the settlement. The compliance provisions cover “the TikTok platform[,]” and “TikTok” is defined as the service “made available to users in the United States under the name ‘TikTok[.]’” The agreement ostensibly does 

 any affiliated products or services, even those that share “user accounts, credentials, login, infrastructure, or content with TikTok.” This seems like it would complicate TikTok’s privacy and data minimization compliance, and it’s an odd match with Alabama releasing the claims it had 

 about TikTok’s “massive data collection” and data sharing in China.

The defendants in the case were TikTok Inc., TikTok LLC, TikTok LTD., TikTok PTE. LTD., TikTok U.S. Data Security Joint Venture LLC, ByteDance, LTD., and ByteDance Inc. The entities released from liability are those entities and every imaginable related entity or individual (Par. 

 lists many, many terms like “indirect subsidiaries, divisions, and joint ventures”). This includes, specifically, TikTok USDS Joint Venture LLC — which, per a footnote, was previously TikTok U.S. Data Security Inc.

The two settlements describe a qualifying

” in nearly identical language as having “at least 5 million U.S. monthly active teens and… an average of at least 30 minutes” daily teen use and permitting them, “as a primary purpose, to create, to share, and to view user-generated videos and images” as well as find “entertainment, social connection, or content discovery of the type that competes for the same teen time spent as TikTok[.]” (I am not sure why this definition is so focused on non-text content.) The definition expressly excludes services for VR, gaming, and “artificial intelligence features, chatbots, or interactions with artificial intelligence.”

In Brazil, the debate over AI development is obsessed with frontier models and data centers, but that obsession has a blind spot. The vast gamut of AI issues goes way beyond that, and for a country that aims to reindustrialize, edge AI, which is when the inference that runs on local hardware rather than in distant data centers, deserves a much louder debate.

The country's biggest AI move so far runs the other way, toward data centralization, supercomputers, and new hyperscale data centers. In August, the Brazilian government 

 it will invest 2.3 billion reais ($444.2 million) in two AI projects, including a supercomputing partnership with China’s Huawei and iFlytek and a tender for a second supercomputers. These supercomputers will run applications built on intensive, centralized data collected and entrusted to the government, powering AI public services such as digital public tools, automated government procurement assistance, and telehealth data transcription. According to the Brazilian Minister of Management and Innovation in Public Services, Esther Dweck, this is a 

strategy dubbed “Base de Dados Brasil” which

 aims to centralize, integrate, and safeguard Brazil's citizens data to be used in AI training. But Brazil is a country of contrasts: it ranks fourth for wealth inequality among the 56 markets in UBS’s 

. Its public investments also contrast, reflecting how resources are distributed.

For example, computer labs are present in only 27% of public schools serving elementary grades, rising to only 73% of public high schools, according to 2024 

 compiled by Todos Pela Educação. A Regional Center for Studies on the Development of the Information Society (Cetic.br) 2025 

 also found that video surveillance cameras are present in 79% of schools in the country's state capitals.

And so, while there's a lively debate over access to Nvidia's latest chips and where to build new data centers, and officials talk about bringing hyperscale infrastructure into the country to pool the data Brazilian citizens generate through public infrastructure, 84% of employed young Brazilians work in jobs that don't require specific qualifications, and only 1 in 7 land a technical or higher-level post, according 

to Brazil's Ministry of Labor and Employment

. Only 12% of high school students want to pursue engineering, and math is a major reason why, 

 by CIEE and Instituto Locomotiva. Young Brazilians rate their confidence with the subject as just 5.2 out of 10, and 79% say gaps in basic education are a barrier to continue their studies. In the Program for International Student Assessment (PISA) 

 published in September, Brazil again scored below the Organization for Economic Cooperation and Development (OECD) average in math.

So, debating only frontier models and hyperscalers is largely buying the vendor's argument dressed up as a development argument. Many public and private sector AI applications in Brazil don't need frontier capabilities. They need reliable, bounded, locally controlled inference — 

 as “edge AI” — and workers ready to fix and maintain this infrastructure.

Why edge AI matters for countries in need of industrial development

Edge AI is a fancy name for a type of data processing realized locally, through small, specialized, well-scoped models that beat chasing frontier scale in industrial data processing, manufacturing, as well as agricultural and lower-tier industrial automation, like defect detection on textile or footwear lines, predictive maintenance in packaging and food processing, or structural monitoring on mining sites in general.

This kind of automation offers a diversified path — provided countries build the standardized, shared policy and funding framework that lets a new economy grow around it, the way China has sought to. China's industrial AI strategy, as mapped in a 

2025 World Economic Forum and Accenture review

, is organized by what each sector actually needs to run. Intelligent transport requires low-latency computing for real-time monitoring and data analysis, while smart manufacturing demands highly precise, specialized computing capabilities. Chinese firms, the report notes, “often find it beneficial to prioritize specialized, smaller models tailored to specific scenarios” than to chase frontier capabilities. The state backs this with layered policy: Beijing sets strategic direction, provinces and cities build AI clusters around their own industrial strengths, and a national program (East Data, West Computing) physically relocates computing capacity toward western regions with abundant renewable energy. This is infrastructure matched to the job.

Brazil doesn't have to invent this model from scratch, but if the country is thinking about reindustrializing by adding a layer of high productivity with AI as the motor, it needs purpose-built edge inference technology, sufficient for the applications that matter most on the ground.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe