Kelsey Piper fed a snippet of her unpublished draft to Anthropic’s Claude Opus 4.7. Just 125 words about binge-watching political TV shows like The West Wing and Servant of the People. The AI nailed it. “The likeliest author is Kelsey Piper,” it declared. No account history. No memory enabled. Incognito mode. Pure prose analysis.
That moment, detailed in The Argument, marks a turning point. AI no longer needs your name or login. It reads your words—your quirks, your tics—and matches them to public writing samples scraped from the web. Piper, a prolific online writer, became the canary in the coal mine. Prolific posters next. Then everyone.
She tested more. A school progress report on a kid’s Pokemon stories. Claude pegged her again. ChatGPT guessed Freddie deBoer; Gemini, Duncan Sabien. A 500-word movie review of To Be or Not to Be. Spot on. Even her dreadful high school college essay. Claude knew. Within exchanges of substance, it knows who it’s talking to.
Anonymity’s shield cracks. Piper cherishes it. As a gay person, she sees it protecting minoritarian lives. It lets outcasts speak. Lets you hear job-losing truths from Nazis, ragebaiters, tankies—sludge she wades through gladly. “Anonymity is the only protection for the outcast,” she writes. Now? Gone for AI chats. Soon, for reviews on Glassdoor. Anonymous posts. All.
But Claude isn’t alone. New research shows AI agents unmasking pseudonymous accounts at scale. A Verge-reported study built systems using unspecified models to scour text for clues. They beat old methods hands down. Link fake profiles to real ones with 67% accuracy, per another report. Your subcultural slang. Mutual friends’ echoes. It all fingerprints you.
Companies pile on. Every major chatbot—ChatGPT, Gemini, Claude, Grok, DeepSeek—trains on user chats by default now. Anthropic flipped in September 2025, the last holdout. A Stanford HAI study confirmed it across six providers. Users expect confidentiality. Fine print grabs broad rights. Data reused for training. Reviewed by humans. Shared across channels.
Browser fingerprinting seals the trap. Screen resolution. Fonts. GPU. Time zone. Harmless alone. Unique together. No cookie needed. Anthropic’s Claude app grabs full system state, memory dumps, browser prints—spyware, some call it on X. Google’s no different. Can’t turn it off.
Industry insiders see the fallout. FTC cracks down on data brokers selling location data tied to clinics, rallies, churches. Mobilewalla. Gravy Analytics. No consent excuses. States like Oregon, Texas demand broker registration. California’s DELETE Act looms, forcing one-click deletions by 2026.
Breaches expose the mess. Sears AI chatbot Samantha leaked 3.7 million chats, addresses, phone numbers—Wired. McDonald’s Olivia bot: 64 million applicant records via weak passwords—another Wired piece. Kids’ AI toy Bondu: 50,000 chat logs open to any Gmail—same outlet. Note app Granola shares links publicly, trains on data unless opted out—Verge PSA.
Regulators push back. UK’s ICO rebukes Google’s IP fingerprinting for ads. EDPB demands consent for tracking pixels. States enforce Global Privacy Control opt-outs. But AI agents roam free—accessing email, Slack, drives. Microsoft’s Recall screenshots everything. Tinder scans your photos. Privacy groups warn Meta’s glasses could ID strangers silently—stalkers’ dream.
Piper predicts adaptation. Write differently. Use AI to rewrite. Unappealing. Local models? Open source offers control—no data harvest, no surprise fine-tunes. But closed labs like OpenAI, Anthropic hold the power: quantize, distill, throttle, block. No changelog. You’re the experiment.
Freedom House notes global anonymity erosion—age verification, weak encryption. 2025 worsened it. Europol sees AI turbocharging dark-web crime: automated attacks, crypto laundering. Yet Vitalik Buterin flags ZK-proofs for pseudonyms, wary of one-ID worlds.
Piper won’t panic-delete her corpus. Embarrassing anyway. She wants awareness. The debate over internet anonymity? Soon anachronistic as iPod Touches. Writers first. Then you. Change styles or lose the mask. AI’s watching your words.
WebProNews is an iEntry Publication