// media

Source: Boing Boing

France's adoption of AI detection technology exposes a familiar vulnerability: governments, like academic institutions, deploy tools with documented high false-positive rates to address legitimate policy problems. When detection science is unreliable, the technology becomes policy theater. Vendors oversell capabilities. Institutions use flawed results to make decisions about student work, hiring, and content moderation. The incentives align poorly.

Source: The Verge

Internal communications show OpenAI and Microsoft explicitly discussed how scaling AI training on web-scraped data would eventually exhaust available training material—and proceeded anyway, treating content depletion as a known externality rather than a problem to solve. This shifts the lawsuit from copyright infringement to deliberate resource extraction: AI companies betting they can monetize publicly-created content faster than creators can replenish it. The court documents amount to a structural admission about how the AI industry plans to sustain itself.

Source: TechCrunch

Microsoft's internal legal filings contradict its public stance on AI training data. The company privately called OpenAI's scraping practices massive intellectual property theft while doing the same to The New York Times. This matters because it shows how AI companies separate their messaging: expressing concern about copyright in court while building trillion-dollar models on uncompensated content. The unsealed documents shift the conversation from abstract fair-use debates to concrete admissions that Silicon Valley insiders view their own practices as indefensible under oath.