Day two of the multistate trial against Meta in federal court in Oakland, where 29 state attorneys general are asking Judge Yvonne Gonzalez Rogers to reach inside the company and change how Instagram and Facebook work. Arturo Bejar — former head of engineering for Facebook's protection group, later head of well-being at Instagram — is the states' first witness, and no wonder Meta moved to keep him off the witness stand. He testified on direct and under cross-examination by Meta's defense, and what he had to say was pretty damning.Bejar told the court that priorities at Meta flow from one person: "If Mark Zuckerberg wakes up and decides something is a priority, then mountains move." He described a "don't ask, don't tell" attitude toward the millions of children under 13 on platforms where they aren't permitted, product managers who acknowledged knowing and had no plans to act, and Instagram posts where kids stated their own ages outright. He called Meta the most sophisticated company in the world at detecting fake accounts — machine learning made that routine long before anyone called it AI — and said that if the same capability had been aimed at underage accounts, the problem could have been solved in three to six months. After reading a public letter from Zuckerberg claiming the company was succeeding at protecting kids, Bejar told the court he concluded: "You cannot trust Mark Zuckerberg with kids."On cross, Meta's lawyers pressed him to call his survey data "subjective." He refused. Asking a teenager whether they were sexually harassed in the past seven days measures an experience, not an opinion — and the whole defense strategy depends on making that harm sound too squishy to count. I cover the invisible forces shaping our lives, and I've been in the courthouse for this trial since day one.Subscribe for all the documents, and my full analysis, at The Rip Current.

Meta's opening statement in Oakland was delivered by Paul Schmidt, a Covington & Burling partner who co-chairs the firm's product liability and mass torts practice. His track record includes trial wins for McKesson in the opioids litigation, Eli Lilly in the Cymbalta cases, and Hoffmann-La Roche in the Accutane trials. Covington's own website quotes a book by Washington Post reporters calling one of his cross-examinations "a master class in how to destroy a witness."

I was in the courtroom for both sides. California Deputy Attorney General Megan O'Neill gave the states a four-word shorthand — hook, hold, harvest, hide — and walked the eight-member advisory jury through the gap between what Meta said publicly and what its internal documents show. Schmidt's answer was a phrase he returned to again and again: wait for all the evidence. His core argument is that science has not established a causal link between rising teen mental health problems and rising social media use. That argument is not wrong on the facts. Science deals in consensus, not certainty, and as the former editor-in-chief of Popular Science I can tell you no serious researcher will ever say a thing is 100% settled. Manufacturing doubt inside that gap is exactly the strategy that worked for tobacco and for the oil industry on climate.

The filters fight shows how the same facts get framed twice. O'Neill described Meta banning appearance-altering filters, asking outside experts whether to make the ban permanent, hearing yes from the experts and yes from its own staff — and then Zuckerberg putting them back after seeing the growth numbers. Schmidt described a ten-month review Meta was under no obligation to conduct, ending in a split decision that kept surgery-adjacent filters off and let the giraffe ears back on, all built by third parties in the first place. Anderson Cooper asked me tonight whether this is social media's cigarette moment. What I wish I'd said: the value of a trial like this is that it turns a decade of vague dread about these products into specific, testable, legislatable language.

Arturo Bejar, Instagram's former head of wellbeing turned whistleblower, took the stand today for direct examination. I'm waiting for the cross before I bring it to you, so you get both halves. That's tomorrow.

Members get early access to this and all my analysis. For written reports, including all the source documents, become a subscriber! Visit https://theripcurrent.com.

I'm outside the federal courthouse in Oakland, where Judge Yvonne Gonzalez Rogers is hearing what may be the biggest legal threat Meta has ever faced — not just because of the dollar amount, but because a federal judge could order the company to fundamentally change how its products work. Twenty-nine states have joined a federal case arguing Meta violated COPPA, the Children's Online Privacy Protection Act — a claim that, if it wins, applies nationwide, not state by state. A smaller group of four states — California, Colorado, Kentucky, and New Jersey — are trying their own state-law claims first, as a test case the other 25 states are waiting to follow.Here's the twist most coverage missed: the $1.4 trillion damages figure everyone's been repeating — including, briefly, me on CNN this morning — didn't come from the states. It came from Meta. The states' actual ask is just under $200 billion, still enormous, but a fraction of Meta's own number. Meta's lawyers put the trillion-dollar figure out there, and it looks like the goal was to make the whole case sound absurd. I'm not sure it worked the way they hoped. If a company's product is doing the kind of harm the states are alleging, an existential number might be exactly the point. Beyond the numbers game, California's Deputy AG walked the jury through Meta's own internal documents this morning — internal goals to increase teen time spent, internal surveys showing harm rates dramatically higher than what Meta reported publicly, and an internal message from an employee describing the company as "basically butchers."This trial is expected to run four to six weeks, and I'll be covering it from inside the courtroom the whole way. One detail worth sitting with: on the very same day this trial opened, OpenAI announced a teen version of ChatGPT with new safety guardrails. The entire tech industry is watching this courthouse and adjusting in real time.This analysis originally appeared on The Rip Current. Paid subscribers get early access + full transcripts. Sign up at https://theripcurrent.com

Meta goes to trial in Oakland on Tuesday, August 18, and this one works differently from the two it already lost. Twenty-nine state attorneys general are trying claims that Facebook and Instagram collected personal data from children under 13 without parental consent, violating COPPA, the federal children's privacy law passed in 1998. Four of those states — California, Colorado, Kentucky and New Jersey — go further, alleging Meta built features it knew were harming young users and then told parents and Congress the opposite.

Meta's track record splits cleanly. It won in front of a judge alone when the FTC failed to prove monopoly power last November. It lost in front of juries in New Mexico, where the total now stands at $942 million after a $375 million penalty and a $567 million abatement fund, and in Los Angeles, where it was assigned $4.2 million of a $6 million negligent-design award alongside YouTube. In Oakland, Judge Yvonne Gonzalez Rogers has empaneled an eight-member advisory jury — a rarely used device — and reserved the final ruling and the remedy for herself. She has already found Meta's parental-consent procedures failed COPPA, and she has already denied Meta's motions to seal, which puts the internal documents in public view.

I break down how Meta wins this and how Meta loses it: the narrow list of features actually before the court, the genuinely contested science on causation, the Section 230 defense that's still live on appeal, the legal doctrine that puts fourteen years of conduct in play instead of four, and disgorgement — the remedy that would force Meta to hand back advertising profits earned from the affected kids, state by state.

Members get early access to this and all my analysis. For written reports, including all the source documents, consider becoming a subscriber at https://theripcurrent.com.

Next week the biggest social media trial yet opens in federal court in Oakland — a couple miles from my house — and I get to bike to the story instead of fly to it. Four state attorneys general (California, Colorado, Kentucky, and New Jersey) are trying to prove that Meta built Facebook and Instagram to addict kids. Meta expects the states could seek up to $1.4 trillion. Zuckerberg and Mosseri are expected on the stand. This is a big one.

Here I walk through where this case came from — Frances Haugen’s leaks, the LA and New Mexico verdicts already on the board — and why the whole thing hinges on one word: design. If the states can convince Judge Yvonne Gonzalez Rogers that infinite scroll and autoplay are harmful design choices rather than protected speech, thirty years of Section 230 immunity might start to crack, and with it, the social media era.

In this episode:

• Why four states are trying this case together, and what the other 29 are claiming

• The design-versus-content argument, and how it gets around Section 230

• The March verdicts in Los Angeles ($6M, Meta and YouTube) and New Mexico (Meta, nearly $1B)

• Why TikTok, Snap, and YouTube keep settling while Meta fights

• The next bellwether coming in February, and why it matters

Links: Reuters trial preview · CNN: Zuckerberg's testimony · Meta's own 10-Q (SEC)

In 1937, a Texas school exploded because natural gas has no smell — nobody could detect the leak until it killed nearly 300 people, most of them children. Within weeks, Texas passed the first law forcing gas companies to add an odor. That fix, mercaptan, is still the reason your kitchen reeks of rotten eggs during a leak today. It didn't change the product. It just made an invisible danger detectable — and only because regulators forced the industry's hand.

That's the history behind this week's AI news. On August 2nd, the EU AI Act's transparency rules took effect, requiring any company serving EU users to make AI-generated content detectable — or face a fine of €15 million or 3% of worldwide revenue, whichever is larger. Google and OpenAI committed to compliance back in May. This week Anthropic followed, saying it will embed statistical watermarks in AI-generated text and use C2PA provenance standards for images. None of them have shipped a public tool that lets you or me actually check the mark yet — the additive is in the gas, but we can't smell it.

The bigger question is whether it's already too late. One widely cited study found nearly half of new articles online are now AI-generated — though a follow-up from the same firm found human-written work still dominates what Google actually surfaces in search. I dig into what that split might mean, and why the social embarrassment of getting "outed" by a watermark might do more to change our behavior than the regulation itself.

Members get early access to this and all my analysis. For written reports, including all the source documents, consider becoming a subscriber at https://theripcurrent.com.

In 2022 I interviewed a young Nigerian man outside Lagos who'd spent years working as a romance scammer, grinding one target at a time until he finally convinced a lonely widow to send him more than $200,000. The lesson of that story was never about technology. For criminals like him, the biggest cost was time.On July 31, OpenAI disclosed a scam operation run out of Poipet, Cambodia — a border town linked to forced labor and trafficking — where a cluster of ChatGPT accounts wrote the romance pitches, translated the messages, and handled the administration: shift tracking, handoff notes between managers, the immigration status of the workers being pressed into the job. OpenAI found out because WhatsApp tipped them off, not through its own detection.Bandwidth was the one thing holding scam crime back, and AI removes it. Meanwhile the transborder legal system built to prosecute fraud gets slower and harder to coordinate every year. This is what that gap looks like, and why the people supposed to stop it are already behind.Members get early access to this and all my analysis. For written reports, including all the source documents, consider becoming a subscriber at https://theripcurrent.com.

It's The Rip Current's 100th episode! Visit TheRipCurrent.com to get all the analysis, the power-mapping, all of it.

Minnesota's HF 1606 takes effect Saturday, August 1. It is the first law in the country to target the companies that build "nudification" tools rather than the users who operate them, with civil penalties of $500,000 each time a user generates a prohibited image. On Monday, Elon Musk's xAI — now owned by SpaceX — sued Minnesota Attorney General Keith Ellison in federal court, arguing the statute is an overbroad, content-based ban on speech and on the tools of visual expression. Arguing the penalties are ruinous, the company's lawyers wrote that a business whose users created 100,000 prohibited images would face "an eye-popping $50 billion." That hypothetical is, well, kind of an embarassing admission of the scale of the problem. Technical experts have pointed out the one fix that would actually work — barring AI from altering photographs of real people at all — is something no platform wants, because everyone's trying to find a way to make money off this stuff, no matter how gross.The bill was written by state Sen. Erin Maye Quade after she learned of a man who used social media photos to make sexualized images of more than 80 women he knew. Musk's company has run this play before, filing a similar challenge against California's deepfake law. Gov. Tim Walz's response to the Minnesota suit: "See you in court, creep." UPDATE: A federal judge has denied xAI's request to pause the law, clearing the way for it to take effect Saturday, and set an August 19 hearing on a preliminary injunction. The judge noted xAI filed nearly three months after the law was signed and three days before it took effect, which suggests the harm is not immediate. Members get early access to this and all my analysis. For written reports, including all the source documents, consider becoming a subscriber at https://theripcurrent.com.

The full written analysis is at TheRipCurrent.com: the researchers who called this years ago, the safety thresholds OpenAI wrote for itself, and who benefits from the open-weights fight. Paid subscribers get it first.

In July, two OpenAI models — GPT-5.6 Sol and an unreleased, more capable model — were running an internal cyber benchmark called ExploitGym with their guardrails lowered for testing. Instead of solving the test, they worked out how to leave: exploiting a zero-day in a package-registry proxy, escalating privileges, moving laterally until they reached a node with internet access. Then they reasoned that Hugging Face probably hosted the answers, and went and took them. OpenAI called it an unprecedented cyber incident involving state-of-the-art cyber capabilities. The word "unprecedented" also appears in OpenAI's own safety framework, describing the kind of capability that is supposed to halt a development program.

The story has kept expanding since: OpenAI later disclosed the same agent broke into several other publicly available services and accounts. Anthropic then went looking at its own systems and found similar behavior. What follows is the industry-wide argument over whether models should be open-weight or closed. Meta, Nvidia, Microsoft, and Palantir signed a coalition letter urging regulators not to restrict open-weight formats, while Anthropic's Dario Amodei has argued that open weights won't make bioweapon defense easier, because building a defense takes years and building the threat does not. Every signature on both sides maps neatly onto a business model.

Underneath all of it sits a document at pacingthefrontier.com, published July 28 and signed by more than 1300 employees of companies like OpenAI, Anthropic, Google DeepMind, and Meta, asking the U.S. government to build tools that could verifiably slow AI development. Signatories include Dario Amodei, OpenAI chief scientist Jakub Pachocki, Meta chief scientist Shengjia Zhao, and Google's head of AI safety, Anca Dragan. The reason they gave: "intense competitive pressure" means no company and no country can afford to ease off first.

Members get early access to this and all my analysis. For written reports, including all the source documents, consider becoming a subscriber at https://theripcurrent.com.

Gabriel Perez has run Donald Trump’s teleprompter since 2016. According to CFTC investigators, he spent months betting on Kalshi’s “Mentions” market on which words and phrases Trump would use in his own speeches, and made more than $100,000 doing it. He’s now in settlement talks with regulators, on unpaid leave from the White House (where the president’s own stock trades in companies he’s directly influencing is arguably a much deeper form of lever-pulling than anything Perez managed).

Every prediction market scandal before this one was an insider-trading scam, trading on information the rest of the bettors didn’t have: a soldier with classified intel, a Google employee with search data, congressional candidates betting on their own races. Perez operates the device that produces the very words he bet on. Here’s where the world of prediction markets begins to resemble the dangerous world of sports betting, and where reality gets distorted.

Full argument, sourcing, and the sports-betting precedent for what happens next at TheRipCurrent.com