ETH Zurich and Anthropic study finds LLMs can unmask anonymous accounts for a few dollars each

Source Cryptopolitan

Researchers at ETH Zurich, AI safety group MATS, and Anthropic have developed an automated LLM pipeline that can connect pseudonymous posts to real-world identities. It was able to identify hundreds of Hacker News users at 90% precision for as low as $1 per user.

The February paper made a comeback this week on X and Reddit.

The paper, “Large-scale online deanonymization with LLMs,” was first posted on arXiv on February 18. Later, it appeared in the proceedings of the 35th USENIX Security Symposium.

The authors are Simon Lermen, Daniel Paleka, Joshua Swanson, Michael Aerni, Nicholas Carlini, and Florian Tramèr. Carlini is an employee of Anthropic, the creator of Claude, and the others are affiliated with ETH Zurich and MATS.

Four stages narrow a pool of 89,000 candidates to a shortlist

No stolen data or hacked server implicated. The agent reads stuff anyone can already see, then does the work of a patient investigator, but more cheaply and faster.

The researchers carved the attack up into four stages, which they called Extract, Search, Reason, and Calibrate. At first, a language model gets identity clues from raw posts, like a hint of a job, a location, or a way of phrasing.

Researchers use LLMs to link anonymous forum accounts to real names for $1 to $4 per target.
Figure 2 of the paper shows the Extract, Search and Reason steps run on an illustrative made-up profile.

Semantic embeddings then narrow a pool of up to 89,000 candidates to a shortlist. A reasoning model grades the best matches and decides if two accounts are the same person.

The Calibrate step causes the pipeline to abstain if not certain. Fewer false positives.

The pipeline runs on off-the-shelf tools, web search, embeddings, and models, e.g., GPT-5.2. Previous deanonymization attacks, such as the 2008 re-identification of the anonymized Netflix ratings, relied on structured data.

The pipeline identified 226 of 338 Hacker News users at 90% precision

To assess the method without risking real people, the team assembled 338 Hacker News users with bios that pointed to a LinkedIn page. Each had a known ground-truth identity.

Fed only comments and submissions, the agent called 226 correctly, with about 67% at 90% precision. The agent got 25 calls wrong and passed on 86 accounts.

Classical non-LLM baselines got close to zero in the paper’s larger matching tests. The experiments cost less than $2,000 in total, or $1 to $4 per profile.

Two other datasets matched Reddit users across two different movie communities and rebuilt one Reddit history by splitting a single history into two time periods. And on each, LLM methods beat the older techniques by a large margin.

The authors cite governments targeting journalists or activists, companies producing ever-sharper ad profiles, and scammers building dossiers for social-engineering pitches.

“The combination is often a unique fingerprint,” Lermen wrote in a February 24 blog post. He added that if a team of smart investigators could identify someone from their posts, LLM agents likely can too, and the cost is only going down.

Researchers did not release the code, prompts, or any real identities turned up by the system. The study was reviewed by the ethics board of ETH Zurich before release.

The paper concludes that “the practical obscurity protecting pseudonymous users online no longer holds.”

The paper’s resurgence comes a day after SEC Commissioner Hester Peirce cautioned against bulk KYC data collection. Peirce said the approach amounts to “building bigger and bigger data haystacks.”

Recent breaches at Revolut and vendors tied to hardware-wallet maker Trezor have fueled fears of “wrench attacks,” in which someone who learns who holds crypto shows up in person.

The smartest crypto minds already read our newsletter. Want in? Join them.

Disclaimer: For information purposes only. Past performance is not indicative of future results.
placeholder
Dollar holds above 100 near a 3-month high — three Fed speakers and a $69 billion auction land tonightThe dollar index closed at 100.43 on Monday, its highest close since late July, after a weekly gain of about 1% — the best in more than three months — and is holding above the 100.00 handle in Asia. Three Fed officials speak tonight alongside a $69 billion two-year note auction, the first leg of $183 billion of Treasury supply this week.
Author  Suzie
Sep 22, Tue
The dollar index closed at 100.43 on Monday, its highest close since late July, after a weekly gain of about 1% — the best in more than three months — and is holding above the 100.00 handle in Asia. Three Fed officials speak tonight alongside a $69 billion two-year note auction, the first leg of $183 billion of Treasury supply this week.
placeholder
October hike odds climb toward 60% as Goldman and BofA both flip — what Warsh's "dose of accommodation" really changedRate futures now price roughly 55% to 62% for a 25bp hike at the October 27-28 FOMC, up from about 30% before Chair Warsh's post-meeting framing that the Fed is merely "removing some accommodation." Goldman Sachs has added an October hike to its forecast and Bank of America now sees moves in both October and December. Here is the repricing, the language behind it, and the two data points that decide it.
Author  Irene Q.
Sep 23, Wed
Rate futures now price roughly 55% to 62% for a 25bp hike at the October 27-28 FOMC, up from about 30% before Chair Warsh's post-meeting framing that the Fed is merely "removing some accommodation." Goldman Sachs has added an October hike to its forecast and Bank of America now sees moves in both October and December. Here is the repricing, the language behind it, and the two data points that decide it.
placeholder
Memory chips surge, Nasdaq notches a second straight record close — why the Dow fell 185 points anywayMicron gained 5%, SanDisk 6.8%, Seagate 4% and Western Digital 3% as the memory complex led the Nasdaq Composite to a second consecutive record close of 27,244.28. But the Dow fell 185 points as JPMorgan, Wells Fargo and Schwab slid more than 3% each — a split tape that says more about positioning than about the economy.
Author  Irene Q.
Sep 23, Wed
Micron gained 5%, SanDisk 6.8%, Seagate 4% and Western Digital 3% as the memory complex led the Nasdaq Composite to a second consecutive record close of 27,244.28. But the Dow fell 185 points as JPMorgan, Wells Fargo and Schwab slid more than 3% each — a split tape that says more about positioning than about the economy.
placeholder
US input costs rose at the fastest pace in four years — the September flash PMI beat is an inflation story, not a growth storyUS September flash PMIs came in far above expectations, with the composite at 58.4, a five-year high. But the detail that moved markets was input cost inflation at its fastest since October 2022, driven by fuel, transport and supply shortages. Brent is back above $100 and the 10-year Treasury yield has hit its highest since 2007.
Author  Suzie
Sep 24, Thu
US September flash PMIs came in far above expectations, with the composite at 58.4, a five-year high. But the detail that moved markets was input cost inflation at its fastest since October 2022, driven by fuel, transport and supply shortages. Brent is back above $100 and the 10-year Treasury yield has hit its highest since 2007.
placeholder
Gold Price Forecast: Gold Drops Below $4,300, Will It Continue to Fall? As of the European session on September 24, gold prices (XAUUSD) extended their correction, dipping below $4,300 intraday to hit a low of $4,262.45. After previously rebounding close to $
Author  TradingKey
Sep 24, Thu
As of the European session on September 24, gold prices (XAUUSD) extended their correction, dipping below $4,300 intraday to hit a low of $4,262.45. After previously rebounding close to $
goTop
quote