Anthropic marks its text as New York moves to curb face scanning
Anthropic plans cryptographic text watermarking while a New York City bill banning biometric recognition in public venues reaches a council majority, moving AI accountability into both product design and law.
Artificial Intelligence··Morning
Anthropic embeds a cryptographic mark in future Claude text
Anthropic says future Claude models will embed an invisible cryptographic watermark in their text, formed by putting a key where word-choice randomness used to sit. A detection API for third parties is on the way; implementation details are still being worked out. The method is a version of Google DeepMind's SynthID-Text: when the model picks among equally valid words, a key replaces the dice, and readability is unchanged. Models released before 2 August 2026 are to receive the watermark on a multi-month schedule. Anthropic writes that the watermark works poorly on small samples, thins out in fact-heavy passages with fewer choices, disappears under end-to-end rewrite, and cannot prove human authorship. In July 2026 the company signed the EU Code of Practice on Transparency of AI-Generated Content with roughly 190 signatories. Accountability arrives first as a product feature in the model output path, with a detection API so the watermark does not live only inside the company. Limits are plain: short samples, fact-dense passages and wholesale rewrites weaken or erase the signal.[1]
A New York bill banning venue face scanning reaches a council majority
A bill proposed in January by New York City Council member Shahana Hanif, banning biometric recognition in public venues, has reached 27 endorsements — more than half the council. Hanif made the case at a press conference outside Madison Square Garden on Friday. Council members are talking with speaker Julie Menin about advancing the bill. The event drew Brooklyn country musician Paisley Fields, a NYC DSA Tech Action representative and members of Fight for the Future. WIRED reported in July that Madison Square Garden keeps a risk-score database for around 400 celebrities and VIPs; the venue sued the magazine for defamation that same month. A venue spokesperson said in a written statement that people have recognised faces since the beginning of time and that MSG simply does it electronically. A council majority does not mean the ban is already law; it shows biometric scanning in public entertainment venues has entered municipal legislation and support has crossed the majority threshold. The dispute continues over whether electronic face matching is ordinary venue practice or unconsented risk scoring.[2]
Accountability enters product design and city law in the same window
The two developments move at different scales and with different tools: one is a cryptographic watermark and third-party detection API in model output, the other a city bill aimed at banning biometric recognition in public venues. On Anthropic's side the limits are written in the company's own text; the watermark weakens on short samples, can vanish under rewrite and does not prove human authorship. On the New York side, 27 council endorsements open a path toward a legal threshold, but the legislative process continues and the ban is not yet in force. The July report on Madison Square Garden's risk-score database and the venue's defamation suit form the dispute's background. Read together, the two tracks show AI accountability advancing as a product decision in the text-generation path and as a municipal bill that could pull face-scanning practice into city law. The shared reader question is who is held to account and by which instrument — an API that reads model output, or a statute that limits biometric devices in public venues. Both answers sit in the same day's agenda; neither founds a final regime yet, but both show accountability moving toward product engineering and local law.[1], [2]
