Fuck-off.ai accuses platforms of selling user data to AI labs, without receipts

An anonymous website called fuck-off.ai circulated on September 19th with a blunt accusation: software companies are expanding their terms so user content can become training material, then presenting customers with a choice between accepting the change and leaving.

The page offers a scenario rather than a disclosed transaction. It says an unnamed frontier AI lab wrote a check large enough to end an unnamed platform's internal debate over licensing user data. The lab, platform, payment, contract and date are all unnamed. Its claim that comparable content deals reach tens or hundreds of millions of dollars also comes without an example.

The essay still lands on a documented fault line. Founders including Mozilla Data Collective's E.M. Lewis-Jong, Kled's Avi Patel, Verb's Cyrus Beschloss and Protege co-founders Bobby Samuels and Travis May are building marketplaces around an alternative premise: if data has become valuable enough to sell to AI developers, the people or institutions producing it should retain control and receive compensation.

That market gives the anonymous page context it cannot supply by itself. It also turns a profane complaint into a practical question for software founders: who owns the value created when customer activity becomes model-training inventory?

The regulatory warning is already on the record

The Federal Trade Commission warned in February 2024 that retroactively adopting broader data practices through quiet changes to terms or privacy policies could be unfair or deceptive. The agency specifically named sharing consumer data with third parties and using it for AI training as practices that may require clearer notice and consent.

A separate FTC warning to AI companies said companies must honor commitments made in privacy policies, marketing materials and product marketplaces. The agency has previously required companies to delete models and algorithms built with unlawfully obtained data, giving the warning consequences beyond a rewritten disclosure.

The anonymous page compresses that legal issue into a familiar product email: terms have changed, continued use counts as acceptance, and the customer's practical remedy is cancellation. Its description is too broad to establish that every platform uses the same defaults or exceptions. Current policies vary by product and customer class.

OpenAI's data-use policy, updated on March 13th, 2026, says content from individual services may be used for training unless the user opts out. OpenAI separately treats business services under different data commitments. Anthropic's July 8th, 2026 privacy update says consumers can control whether conversations improve its models and excludes Team, Enterprise and developer-platform accounts from those consumer terms.

Those distinctions matter to founders selling collaboration tools, developer platforms and AI products. A privacy promise can operate as a product feature for enterprise buyers while consumer data remains an input elsewhere in the same organization. The policy may be legally segmented. Customers still experience the company as one brand.

Founders are building a paid alternative

E.M. Lewis-Jong, Mozilla Data Collective's founder and CEO, came to the problem after leading Mozilla's Common Voice project, a crowdsourced speech corpus covering hundreds of languages. Mozilla says Lewis-Jong started the collective in 2025 after seeking a platform that would give data-producing communities control over licenses and value exchange.

Mozilla Data Collective launched Compensated Datasets on July 30th. Verified providers can set licensing prices and receive the license fee, while Mozilla charges buyers a separate 5% platform fee. The structure makes permission, price and intended use explicit before a dataset changes hands.

Avi Patel is pursuing the consumer version at Kled. Kled announced a $5.5M seed round on March 10th, bringing its stated total financing to $10M. The round included Wischoff VC, Aglae, K5 Global, Parable VC and Cox Exponential, alongside individual investors. Kled says contributors upload material for licensed datasets and can receive payment when that data is purchased. Its upload, earnings and customer figures remain company-reported.

Cyrus Beschloss, Verb's co-founder and CEO, took a similarly direct approach with a marketplace launched in August. Verb lets consumers select categories of phone activity to collect, set prices and block sales to chosen buyers. The model asks users to trust another intermediary with sensitive behavioral data, leaving security and buyer oversight as central product risks.

Protege is working farther up the supply chain. Bobby Samuels and Travis May founded the New York company to license proprietary datasets held by organizations including healthcare and media groups. Protege announced a $30M financing led by Andreessen Horowitz in January, expanding its August 2025 Series A and bringing its stated total funding to $65M. Footwork, CRV, Bloomberg Beta, Flex Capital and Shaper Capital returned as investors.

These founders are betting that provenance, permission and compensation will become infrastructure rather than public-relations language. AI developers need differentiated data, and data holders increasingly understand that access has a price. A marketplace can document who supplied a dataset, which rights travel with it and how revenue is divided.

The receipt is the story

Fuck-off.ai identifies the incentive cleanly: platforms can treat user activity as an asset once AI labs are willing to pay for access. The site's sweeping presentation weakens its own case because the alleged transaction cannot be examined. There is no way to test whether users consented, whether personal information was transferred, whether content was aggregated, or whether the contract involved training at all.

The stronger version of the argument is visible in the companies forming around paid data exchange. Their founders are attaching prices, controls and contracts to material that platforms historically collected as a condition of access. They also inherit a harder burden. Consent must remain understandable after the onboarding screen, compensation must reach the people promised a share, and provenance claims must survive contact with an enterprise buyer's legal team.

For software founders, the immediate decision is architectural as much as legal. Products built around customer content need explicit answers about whether that content trains internal models, reaches outside model providers, survives an opt-out, or changes treatment across free, paid and enterprise accounts. Hiding those answers inside a policy update creates a trust debt that no carefully drafted acceptance clause can erase.