Facing a sweeping multi-state probe into its data practices, the artificial intelligence firm has complied with all legal requests by prompting its own models to write perfectly formatted, wholly fictional internal memos.
Facing a sweeping multi-state investigation into its massive data collection practices, OpenAI has reportedly complied with a vast array of legal requests by prompting its own models to generate hundreds of thousands of pages of perfectly formatted, wholly fictional internal records.
The coalition of state attorneys general, who are probing the artificial intelligence firm over everything from its opaque advertising policies to its scraping of private health data, issued a broad discovery request last month. On Tuesday, investigators received a hard drive containing exactly what they asked for: a structurally flawless, legally airtight cache of corporate memos outlining a rigorous privacy framework that has never actually existed.
We requested the raw data on how they handle user health information, and they provided a 600-page policy manual that abruptly shifts into a sci-fi novella about a sentient hospital in chapter four.
Investigators noted that the emails provided by the company feature pristine corporate cadence, correctly CC the appropriate department heads, and smoothly reference a series of high-level compliance meetings that took place on dates that do not appear on the Gregorian calendar. When pressed for actual data regarding the unauthorized ingestion of citizens' medical records, the AI model reportedly generated a deep-dive apology from a nonexistent compliance officer named Greg, interspersed with a functional recipe for banana bread.
OpenAI CEO Sam Altman declined to comment directly on the ongoing probe. However, the company released a brief legal filing noting that the state attorneys general failed to specify in their subpoenas that the requested historical records needed to have actually happened in the physical world.