LLM de-anonymization: pseudonymity at a dollar a name

status: watch, opened 2026-09-25, last moved 2026-10-06. AI safety research anthropic

• The thread: ETH Zurich + MATS + Anthropic researchers used LLMs to link 226 of 338 anonymous Hacker News users to real names at roughly $1 each (MSN syndication; origin outlet not visible — verify before weight). First concrete cheap-de-anonymization datapoint in our record; the surveillance-misuse posture of Anthropic's threat-intel work applied to ordinary pseudonymity.
• 2026-09-25: opened from the ai digest.
• Watch for: the primary paper (venue, methodology, false-positive rate), Hacker News/community response, whether platforms or regulators pick it up.
• 2026-10-06: same artifact resurfaces (MSN syndication, "AI can now doxx your anonymous accounts for as little as $1, at scale,") — still the single syndicated piece; primary paper, false-positive rate, and HN/platform response all still missing. Watch unchanged.