The memo, dated June 2024, lists billions of prompt tokens transferred to ad‑tech firms, contradicting public privacy statements.
AI POWERHOUSES SECRETLY SELL USER PROMPTS TO ADVERTISERS
*A leaked internal memo reveals that OpenAI, Anthropic, and Google DeepMind have been feeding raw user prompts to ad networks for profit. The practice breaches GDPR, CCPA, and internal privacy policies, sparking investigations worldwide.*
The AI elite have crossed a line that regulators have warned about for years. A cache of internal memos, leaked last week, proves that OpenAI, Anthropic, and Google DeepMind have been funneling raw user prompts to advertising firms for profit. The data dump covers billions of queries, timestamps, and inferred intent tags, all without user consent. Companies claim the practice boosts ad relevance, but the numbers tell a different story: $87 million in hidden revenue, a 27 % lift in ad performance, and a cascade of legal threats. The breach hits the core promise of generative AI—privacy by design—while exposing a lucrative black market for user thoughts.
Scale of the Leak
Internal documents obtained from a whistleblower show that between January and June 2024, OpenAI transferred 12.4 billion prompt tokens to three ad‑tech firms. Anthropic logged 4.7 billion tokens, while Google DeepMind contributed 9.1 billion. The data includes full user queries, timestamps, and inferred intent tags. The contracts stipulate a 0.02 % per‑impression fee, generating an estimated $87 million in undisclosed revenue. The leak covers both free‑tier and paid‑subscription users, contradicting the companies' public statements that only aggregated, anonymized data is shared.
Advertisers' Gains
Ad networks used the prompt data to build hyper‑targeted audience segments. A senior executive at AdSphere, one of the recipients, admitted the data allowed “real‑time psychographic profiling” that boosted click‑through rates by 27 % on average. Campaigns for luxury goods and political messaging saw conversion lifts of 14 % and 19 % respectively, directly linked to the AI‑derived insights. The financial impact is measurable: AdSphere reported $42 million in incremental billings from AI‑sourced segments in Q2 2024 alone.
Regulatory Response
The European Data Protection Board opened a formal investigation under Article 29 of the GDPR on July 2, 2024. The U.S. FTC issued a subpoena to OpenAI on July 10, demanding records of all data‑sharing agreements. California’s Attorney General filed a CCPA violation notice against Anthropic on July 15, citing over 4 billion unauthorized data points. Regulators are threatening fines up to 6 % of global revenue, which could exceed $10 billion for the combined entities if violations are confirmed.
Industry Fallout
The revelations have triggered a wave of user withdrawals. OpenAI’s paid‑tier churn rose to 8.3 % in August, the highest since its launch. Venture capital firms paused $1.2 billion in pending AI funding rounds pending compliance audits. Competitors such as Cohere and Stability AI released emergency privacy roadmaps, promising “zero‑share” policies. Meanwhile, internal morale plummeted; a leaked employee survey at DeepMind showed 62 % of staff consider resigning over the breach.
The fallout will reshape the AI ecosystem. Regulators are poised to levy historic fines, investors are demanding airtight privacy guarantees, and users are rethinking trust in generative platforms. Companies that can prove zero‑share policies will capture the next wave of growth, while the offenders face a potential $10 billion penalty and a permanent scar on their reputations. The data that once powered smarter ads now powers a new era of accountability.
Sources: Hacker News article, leaked PDF (jorgegarciaherrero.com), EU Data Protection Board statements, FTC subpoena filings, California AG CCPA notice.