OPENAI PRIVACY POLICY BUSINESS SERVICES AGREEMENT META TERMS OF SERVICE ANTHROPIC TRANSPARENCY HUB SOMANTIX SCRAPE FAILURE REPORT MISC MULTI

The Rest

Vol26, Issue 13 Smaller Changes

OpenAI added one new opt-out path inside its cross-context behavioral advertising disclosure paragraph this week. Logged-out users can now opt out “within Settings > Data Controls on ChatGPT or using the Your Privacy Choices link on our website.” The “Settings > Data Controls on ChatGPT” path is new — meaning OpenAI has now built an in-product opt-out mechanism inside ChatGPT itself and surfaced it in the same paragraph where the ad-data sharing is disclosed. This is the third consecutive month that paragraph has been touched, and the third consecutive month the touch has been protective scaffolding around the ad infrastructure rather than the infrastructure itself.

The contract’s reference to the ChatGPT Pricing Page changed domains: prior version pointed to openai.com/chatgpt/pricing; updated version points to chatgpt.com/pricing (with referral and tracking parameters appended). Small as a diff, but the domain change is a beat in OpenAI’s ongoing migration of the ChatGPT consumer surface off the openai.com domain. Worth tracking as a thread

Now in tracking for the first time after several silent scraper failures. The document carries a release date of April 29, 2025 — so this is a baseline being established, not a new agreement. Notable provisions for the audience: a $1,000 minimum liability cap, an explicit “Meta will not use your Content to train any artificial intelligence models” commitment, a 700-million-monthly-active-users commercial restriction, an EU prohibition on accessing multimodal AI models via the API, and explicit Illinois BIPA / Texas CUBI warranties in the user representations. Will be tracked normally from this issue forward.

Anthropic Anthropic Transparency Hub

One row in the Claude 3.7 Sonnet model card changed from a list of access surfaces to “Claude 3.7 Sonnet is retired and no longer available for use.” Operational rather than contractual.

Pond5 License Agreement, Shutterstock License Agreement, Shutterstock Privacy Policy, Shutterstock Terms of Service. All four returned the identical 43-byte anti-bot challenge string — “Please enable JS and disable any ad blocker.” The JS-rendering bypass flagged in the Vol. 26 No. 20 handoff has not yet shipped. Until it does, four of the most heavily-used licensing documents in the stock-media supply chain are currently invisible to this newsletter.

Misc Misc Multi

Anthropic (Privacy Policy and Consumer Health Data Policy — URL redirect token rotation in both); Autodesk (page chrome footer text change); Cascadeur (language picker and reCAPTCHA scrape artifacts reshuffled); Google Gemini API (language picker URLs stripped to plain text, “On this page” TOC removed); Hugging Face (StripeM-Inner footer artifact added); LangChain Shared Responsibility Model (Interrupt conference banner removed, LLM Gateway promoted from inline tag to subsection); LightTricks (four blank lines added at end of a 1.3 MB document); Meta (Privacy Policy and Terms of Service — URL &h= tracking token rotation in both); N8N (entire Cloudflare Turnstile UI captured as scrape artifact); Synthesia (author byline expanded from “Meg” to “Meg Farley” — last week’s flag resolved). Same pattern, different documents, no policy text changed.

This originally appeared in Vol. 26, No. 13, ComfyUI: The Pixels Are Yours, The Recipe Isn't

View Full Issue →

Sign up for The Ledger Newsletter