Andreessen Horowitz is shaping AI policy — while investing in a bleak vision of the future - Model Republic

Marc Andreessen wants to shape US AI policy. The venture capital firm he co-founded and runs, Andreessen Horowitz (abbreviated “a16z”), is a major player in the development of new tech startups. These startups include:A bot farm of fake accounts, tricking people and social media platforms into thinking AI-generated ads are posted by real peopleAn AI company that wants to normalize cheating on dates, job interviews, and tests with AIAI companion apps linked to suicide and disturbing behavior toward childrenA platform hosting thousands of deepfake models — 96% targeting identifiable women — that have been used to create AI-generated content sexualizing childrenGambling platforms that attempt to subvert existing laws and target vulnerable usersFintech companies implicated in fraud and illegalityMany of these companies knew the rules and broke them anyway — or designed products specifically to exploit gaps in consumer protection. The firms profited, and the public paid the costs.There’s a growing public desire to rein in tech companies and regulate AI, so a16z is spending tens of millions of dollars to shape the development of AI policy. The firm helped launch a $100 million super PAC, saw former partners take key government roles, and successfully pushed for an executive order attempting to undermine state AI laws. The partners want to set the rules of the road, even as they’re already driving recklessly.What follows is The Midas Project's survey of 18 of Andreessen Horowitz's most notorious investments. This isn’t a comprehensive overview of the firm’s larger portfolio, but it indicates a pattern of behavior — one comprising hundreds of millions of dollars of investment by a16z.These investments reveal the lines that a16z is willing to cross and how the lax regulatory environment that they favor would benefit the firm’s bottom line.A16z did not respond to a request to comment for this report.Deception and manipulationA16z has invested in products designed for mass deception. Even if these tactics don’t explicitly violate the law, they can be corrosive to society.As technology like advanced AI improves — making it much easier to fake almost anything — decision makers may want to enact new laws or policies that mitigate the social costs. And if a16z gets its way, we might never update the rulebook.DoublespeedA16z invested $1 million in October 2025 via Speedrun.Doublespeed sells the capacity to trick everyday people, and social media platforms themselves, into thinking AI-generated ads are genuine human content. Here are some select quotes from the company’s promotional video:“We run the only VC-backed bot farm in America. Because why let Russia and China have all the fun?” “We didn't break the internet. It was broken to begin with. But now we're killing it entirely.” “Welcome to the dead internet.”A16z's Speedrun program invested $1 million in Doublespeed, a company that was recently covered in a blistering article by 404 Media, which reported: “Andreessen Horowitz is funding a company that clearly violates the inauthentic behavior policies of every major social media platform.”Excerpts from Doublespeed’s websiteThe company’s business model relies on deception, designed to make social media platforms and their users believe AI-generated images and videos depict real people. How do they do this? By selling access to “phone farms” that create and manage thousands of fake social media accounts to manipulate engagement metrics. The company's website is explicit, saying its product “mimics” the behavior of real people on social media in order to “get our content to appear human to the algorithms.”“Yes, we built a phone farm (and its pretty sick),” said Doublespeed founder Zuhair Lakhani on X. The purpose was “replacing human creators with ai, mainly used for marketing.” A photo of Doublespeed’s phone farms, shared by the founder Zuhair Lakhani on X.They use thousands of real phones to pull this off because social media platforms like TikTok have policies against and methods to detect the mass generation and deployment of fake accounts. The company has the accounts imitate human behavior before posting deceptive content. This means the fake accounts search specific keywords, scroll their “For You” pages, and use AI to analyze screenshots of content to determine whether to “repost it, comment on it” or “swipe away.” A feed of AI-generated marketing content created by Doublespeed. Source: Superwall on YouTubeA selection of nearly identical Doublespeed-run TikTok accounts. Most posts involve the AI decoy complaining about any one of a number of medical issues. Then, the account lists a handful of cures, including a foam roller product from Doublespeed’s client. Source: Tiktok, Doublespeed on loomThis is all designed to circumvent platforms’ restrictions on fake content and then serve that fake content to unsuspecting real people.In a podcast interview, Lakhani offered details about one of the company’s clients: “They're hitting like the old person niche, which is what I think is like the best niche to hit with AI content.” Polling and research have found that older people are less likely to say they’ve heard about AI and more likely to fall for AI-generated misinformation.Lakhani drew a parallel between this client and his prior work producing AI-generated marketing content at scale: “It was all like old person niche stuff. So like all supplements that would, you know, target old people, and that's when the commission would go crazy.”“Those brands would tell you to do like, you know, make some like extremely crazy claims,” he said, “especially with supplements.” Lakhani added, “The supplement stuff should definitely be like kind of illegal. I don't know how that is allowed.”Despite their founder stating that supplement ads should be illegal, Doublespeed isn’t shying away from them. In December 2025, a hacker gained access to Doublespeed’s entire backend and the leaked data showed what the AI-generated “influencers” were actually selling. One account, “pattyluvslife,” featured an AI-generated woman claiming to be a UCLA student. The account criticized the supplement industry and pharmaceutical companies as fraudulent — while simultaneously promoting a herbal supplement from a brand called Rosabella.Another account under the name “chloedav1s_” had uploaded some 200 posts featuring an AI-generated woman claiming to suffer from various health conditions and often pictured in a hospital bed. She ultimately promoted a specific company’s foam roller as a solution to her ailments.A tweet from DoubleSpeed’s founder shows one of the company’s bot accounts messaging a user to promote the product. In the post, Lakhani boasted, “A couple of weeks ago, we gave the [AI] agents access to dm … This was for an ecommerce brand - out of 130 dms sent, 15 pointed to a conversion.”Another image from Doublespeed’s platform showing their bot account, imitating a human and messaging users with medical conditions to promote the client’s foam roller product. Source: Zuhair Lakhani on X.The Doublespeed hack revealed more than 1,100 phones and over 400 TikTok accounts operated by the company. Most of the accounts were promoting products without disclosing that the posts were paid advertisements — a violation of both TikTok's Community Guidelines, which require creators to label AI-generated content depicting realistic scenes, and FTC regulations, which require influencers to clearly disclose any “material connection” to a brand when endorsing products.Doublespeed and a16z did not respond to 404 Media’s requests for comment. After 404 Media flagged the accounts to TikTok, the platform said it added labels indicating they were AI-generated. However, The Midas Project’s follow-up investigation has revealed that while labels have been added to some content from some Doublespeed-run accounts (including chloedav1s_), others with comparable reach and near-identical content still remain unlabeled (such as lilyw4tson and mia.garc1a), with most commenters appearing to believe the posts are authentic.Cluely AIA16z led a $15 million Series A in June 2025.Cluely's official manifesto declares: “We want to cheat on everything. Yep, you heard that right. Sales calls. Meetings. Negotiations. If there's a faster way to win — we'll take it... So, start cheating. Because when everyone does, no one is.”Cluely’s co-founders Neel Shanmugam (left), Roy Lee (center), and Alex Chen (right). Source: Cluely via Bloomberg.Founder and CEO Roy Lee is no stranger to using AI to cheat. By his own admission to New York Magazine, while studying at Columbia, he used AI to cheat on “nearly every assignment,” estimating that ChatGPT wrote 80% of every essay he turned in. “At the end, I'd put on the finishing touches. I'd just insert 20 percent of my humanity, my voice, into it.”In early 2025, Lee built Interview Coder, a tool that operates behind-the-scenes during technical coding interviews and feeds AI-generated solutions to users in real time. He recorded himself using it to pass Amazon's interview, received a job offer, publicly declined it with mockery, and posted the video to YouTube. He also claimed to receive offers from TikTok, Meta, and Capital One. Amazon reported him to Columbia. The university placed him on probation for “facilitation of academic dishonesty.”“Even if I say extremely crazy shit online,” Lee has explained, “it will just make more people interested in me and the company and it will just drive more downloads and conversions and get more eyeballs onto Cluely.”A marketing video for Cluely suggests that the product can be used discreetly to “cheat” on dates. Source: YouTubeCluely's launch video demonstrated another of the product's intended use cases: dating. In it, Lee goes on a blind date and uses the tool to lie about his age, job, and interests. It has so far amassed 13 million views on X.Under scrutiny, Cluely has quietly walked back some of its original