AirTag reveals Amazon is trashing rare books to train AI
THE SO WHAT
Physically shredding rare books to digitize them for AI training collapses the gap between data hunger and cultural preservation — and it will harden regulatory and public scrutiny around training corpora. Any operator using physical or archival content for model training should assume provenance, consent, and preservation practices will be audited, not ignored.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAI automation startup Relay shuts down, staff joins Google’s Chrome team
Talent and IP from niche AI workflow tools are getting absorbed into core platform teams—Chrome picking up Relay’s people means browser surfaces will keep pulling more automation natively into the stack. If you’re building horizontal AI automation, assume the browser, OS, and office suite are on your roadmap as competitors, not just integration points.
Applied AIOpenAI disbanded the team that assessed catastrophic model risks
Folding catastrophic risk work into existing leadership lines concentrates both capability and accountability—governance is moving from a standalone function to an embedded constraint on product and research. For enterprises depending on frontier models, vendor due diligence now needs to ask where long‑horizon risk lives on the org chart and how it’s resourced, not just whether a team exists on paper.
Applied AIOpenAI is funding 13 Axios local newsletters in a three-year deal
OpenAI underwriting 13 Axios Local newsletters in exchange for training rights is the template: AI labs fund journalism and infra, publishers trade fresh data and adoption. If you own proprietary content, you’re now negotiating both cash and model access—decide what your archive is worth before a lab sets the anchor.
Google is buying Spirit Airlines' old data for $10 million to improve its AI
Training data is now an asset class — $10 million for a single airline’s historical data shows how aggressively large players are sweeping up proprietary corpora. If your business sits on unique behavioral or operational data, treat it like IP and price it accordingly before someone else does.