What Meta Knew and When

A slide that Meta's researchers wrote, for Meta's leadership, in Meta's template, read: "We make body image issues worse for one in three teen girls." A second slide, drawn from the same internal research, read: "Teens blame Instagram for increases in the rate of anxiety and depression. This reaction was unprompted and consistent across all groups." A third reported that, of the teen girls in the United Kingdom who had experienced suicidal thoughts, 13 percent traced those thoughts back to Instagram. The U.S. figure was 6 percent.

These sentences were not written by a critic of the company. They were not the product of an outside study Meta could dispute. They were the company's own findings, in the company's own document templates, presented in 2019 to senior leadership including Adam Mosseri, the head of Instagram. They became public two years later because a single employee took photographs of them with her phone (Wells, Horwitz, and Seetharaman, Wall Street Journal, September 14, 2021).

Her name was Frances Haugen. She had joined Facebook in June 2019, after more than a decade as a product manager at companies including Google, Pinterest, and Yelp, where she specialized in algorithmic recommendation systems (Haugen testimony, U.S. Senate Subcommittee on Consumer Protection, October 5, 2021). Facebook recruited her for that expertise. Her team was assigned to study how the platform's recommendation systems fed political polarization and misinformation ahead of the 2020 U.S. presidential election.

The civic integrity team had been formed in 2018, in response to internal concerns about the 2016 election. It was dissolved in December 2020, soon after the next election was decided. The research, which Haugen described as some of the most rigorous internal social-science work she had ever encountered, was parceled out to product groups whose mandates had nothing to do with the questions her team had been built to answer (Haugen testimony, U.S. Senate Subcommittee on Consumer Protection, October 5, 2021).

So she copied what she could. Through the spring of 2021 she photographed her screen while logged into Workplace, Facebook's internal collaboration platform. She captured material on Instagram's effects on teen mental health, content amplification, misinformation, and user well-being. She left in May. By late summer she had retained an attorney and filed eight separate complaints with the Securities and Exchange Commission, alleging that Facebook's public statements to investors had materially misrepresented what its own internal research had found (Securities and Exchange Commission complaints, Haugen et al., August 2021).

She also gave a portion of the documents to reporters at The Wall Street Journal.

The Facebook Files

On September 13, 2021, the Journal published the first article in what came to be called the Facebook Files. The lead piece described a Facebook program called Cross Check that exempted high-profile accounts from the platform's normal content-moderation rules. By 2020 the program covered roughly 5.8 million users, creating a two-tier system at odds with what the company was telling the public about how moderation worked (Horwitz, Wall Street Journal, September 13, 2021).

The next day brought the piece that would define the rest of the series: Facebook Knows Instagram Is Toxic for Teen Girls, Company Documents Show, by Georgia Wells, Jeff Horwitz, and Deepa Seetharaman (Wells, Horwitz, and Seetharaman, Wall Street Journal, September 14, 2021). The slides quoted at the top of this chapter were the ones the Journal published. The article walked through them one by one.

The series ran through the fall. A third article reported on internal research showing that Facebook's 2018 algorithm change, sold publicly as a shift toward "meaningful social interactions," had amplified divisive, outrage-driven content. That content drew the most engagement (Hagey and Horwitz, Wall Street Journal, September 15, 2021). A fourth reported that Facebook employees had flagged the platform's use by drug cartels and human traffickers, including the sale of domestic workers, and that the company's response had been weak (Scheck, Purnell, and Horwitz, Wall Street Journal, September 16, 2021). On October 5, Haugen testified before the Senate, identifying herself as the source of the documents and answering questions under oath.

The company did not dispute that the documents were real. It disputed how they were being read. Adam Mosseri acknowledged the internal research, noted that the same research showed positive effects on the mental health of many teen girls as well as negative ones, and said the company was working to make the platform safer (Mosseri, NBC's TODAY, September 27, 2021). Meta also paused development of Instagram Kids, a separate platform intended for users under thirteen (Mosseri, Instagram Newsroom, September 27, 2021).

By the end of October the news cycle had moved on. The share price had dipped and steadied. No executive resigned. The civic integrity team was not reconstituted. The Facebook Files entered the public record. They did not change how the company operated.

The October 2023 filing

On October 24, 2023, a coordinated lawsuit was filed in the United States District Court for the Northern District of California, brought by thirty-three states and the District of Columbia against Meta Platforms and its subsidiary Instagram (People of the State of California et al. v. Meta Platforms, Inc., Northern District of California, October 24, 2023). The total eventually grew to 41 attorneys general plus D.C.

The 233-page complaint drew on documents obtained through a coordinated multi-state investigation that had opened in 2021, shortly after the Facebook Files. The investigators had used the subpoena authority of the participating state attorneys general to pull internal company documents the 2021 disclosures had not reached. Several of those documents appear in the complaint by direct quotation.

Meta's own researchers had found that a majority of teen girls experience negative social comparison on Instagram, and that a significant share of them believed the app made it worse (Complaint, paragraph 203). In internal emails in February 2021, employees debated whether the company even wanted to reduce that effect. Social comparison was, as the discussion put it, good for the business and bad for teen girls. "Do we want social comparison or not?" one asked (Complaint, paragraph 229). Elsewhere the complaint alleges that Meta represented its platforms as safe while its own data showed users experiencing harm at far higher rates (Complaint, paragraphs 458 through 507).

The complaint alleged that Meta publicly represented its platforms as safe for users aged thirteen and older while its own records showed millions of children under thirteen using Instagram, by the company's count at one point roughly a third of all ten to twelve year olds in the country. Those accounts violated Meta's own terms of service. Keeping them, the complaint alleged, was not an enforcement lapse but a choice made for revenue reasons (Complaint, paragraphs 642 through 811). In 2021, Mark Zuckerberg told Congress that Meta removes users under thirteen.

Meta has denied the substantive allegations. The case is, as of this writing, in active litigation.

How the information moves

What allows a multi-year disclosure arc like this one is not the absence of internal awareness inside the company. It is structure. The company compartmentalizes what it knows, routes it through legal privilege, and keeps it from ever forcing a response that would change the underlying business model.

Three features matter. First: legal privilege. User-experience research conducted under the supervision of in-house counsel and circulated through privileged channels can be designated as work product prepared in anticipation of litigation. Material like that is generally not discoverable in subsequent legal proceedings unless a court finds the privilege has been waived. The practical effect: researchers can be candid in a way they could not afford to be if the work were freely discoverable.

Second: team-level compartmentalization. Internal findings on adolescent harm were, by Haugen's testimony and by what the attorneys general later obtained, circulated within Meta on a need-to-know basis. A senior executive could be honest with a particular research team about a finding that the same executive would not have included in a public investor communication.

Third: the settlement-and-non-disclosure layer. When internal disputes between the company and its researchers reach formal escalation, the resulting settlements often include non-disclosure terms that keep the underlying findings out of public view. Researchers who leave under those terms cannot publish the findings, even when the work falls well within their own expertise.

None of this is unique to Meta. This is standard practice for any large American enterprise that expects serious exposure to civil litigation. The pattern is older than the industry that currently displays it. The tobacco companies ran the same play for decades. Their own scientists documented that nicotine was addictive and that smoking caused disease while the companies told the public the science was unsettled, and that gap between the internal record and the public position became a matter of law only when the documents were forced into the open through the litigation that produced the 1998 Master Settlement Agreement (Tobacco Master Settlement Agreement, 1998).

What the record does and does not show

The 2021 and 2023 disclosures establish four things: that Meta's internal research found harm to a meaningful subset of its adolescent users; that the company's product teams were aware of the research; that the company's public statements to investors and regulators did not fully describe the findings; and that what the public was told about adolescent safety did not match what the company's researchers were telling its leadership.

What the disclosures do not establish, at the level of legal proof, is what individual executives knew at specific moments, or what their personal motivations were. If the lawsuits proceed to discovery and trial, they will test those questions under oath. Those questions have not been answered yet.

There is a scientific limit on what these documents prove. The strongest internal findings, including the body-image research the September 2021 article cited, concern specific subgroups of adolescent users in specific use patterns. They do not establish that Instagram or Facebook causes population-wide harm at any fixed level.

Federal law adds further constraints on what regulators can do with disclosures of this kind. Section 230 of the Communications Decency Act of 1996 provides broad immunity to platforms for content posted by users. The First Amendment further limits the federal government's authority to regulate platform features that are arguably expressive in character. Recent state-level efforts to regulate adolescent social-media use, such as Utah's 2023 Social Media Regulation Act and Florida's 2024 HB 3, are in active litigation against platform-industry plaintiffs.

That legal terrain is not the subject of this chapter. The gap between what the record shows and what regulators have done about it is not inattention. The constraints are real. A later chapter examines them directly.

The pattern this chapter has traced is documentary, not moral. The conclusions a reader draws are the reader's. The record is what it is.

Who kept funding it

The record shows what Meta knew and roughly when it knew it. Given how little regulators did with that record, the natural question is who kept the money flowing.

The Facebook Files did not sink the company. Meta's stock did fall hard in 2022, including a single day that February when it lost about 232 billion dollars in value, the largest one-day loss in American market history to that point. But that collapse was driven by an advertising slowdown, Apple's privacy changes, and the cost of the company's hardware bets, not by the disclosures (CNBC, February 3, 2022). By late 2022 the company had shed roughly two-thirds of its peak value. Then it recovered. Meta's market value sat near one trillion dollars when the documents broke in 2021. By early 2026 it was worth around 1.5 trillion, well above where it stood when the disclosures began (CompaniesMarketCap, 2026; Macrotrends, 2026).

Ordinary investors financed that recovery. If you hold a retirement account, the odds are that you were one of them.