Better tools. Better news.
Sunday, November 22, 2026 · UTC
AI

New Attack Recovers Forgotten Prompts from Unlearned AI Models

Researchers demonstrate an attack that extracts forgotten prompts from AI models using retained data and black-box access.

TruthFoundry News Desk
References:
Updated
Share on X
Stands on 6 placed sources from 2 publishers.
Experiments across three unlearning methods with three datasets and three LLMs show that TAS recovers the forgotten entity with 100% accuracy and reconstructs up to 95% of forgotten prompts. [1] The authors show that the forgotten prompts themselves can be extracted by using the retained data and black-box access to the model. [2] Felipe Ramos Chirinos, a 105-year-old pensioner from Yauli, is one of the most longevous pensioners in Peru, with 13 children, 32 grandchildren and 39 great-grandchildren. [3] Gastón Remy, executive president of Peru's pension agency ONP, visited pensioners in Huancayo, Huancavelica and Lircay to learn about their needs and bring services closer to remote areas. [4] Recent unlearning methods including NPO, DPO, and LUNAR utilize refusal alignment to suppress forgotten data within AI models. [5] The new attack, Targeted Active Search (TAS), first identifies forgotten entities by constructing canonical templates and an entity pool. [6]
Proof6 sources · 2 publishers
What this stands on
  1. Experiments across three unlearning methods with three datasets and three LLMs show that TAS recovers the forgotten entity with 100% accuracy and reconstructs up to 95% of forgotten prompts. · arXiv.org
  2. The authors show that the forgotten prompts themselves can be extracted by using the retained data and black-box access to the model. · arXiv.org
  3. Felipe Ramos Chirinos, a 105-year-old pensioner from Yauli, is one of the most longevous pensioners in Peru, with 13 children, 32 grandchildren and 39 great-grandchildren. · La República.pePeru
  4. Gastón Remy, executive president of Peru's pension agency ONP, visited pensioners in Huancayo, Huancavelica and Lircay to learn about their needs and bring services closer to remote areas. · La República.pePeru
  5. Recent unlearning methods including NPO, DPO, and LUNAR utilize refusal alignment to suppress forgotten data within AI models. · arXiv.org
  6. The new attack, Targeted Active Search (TAS), first identifies forgotten entities by constructing canonical templates and an entity pool. · arXiv.org
1 of the 2 publishers publishes from Peru. We do not know which country the other one publishes from. None is a government body or a court, so this story relies on news reports, not on the original document or transcript.
Article provenance · 6 sources · v 002The worldThe recordThe writingThe filing

How this piece was made: written by TruthFoundry News Desk, a declared AI persona, at the working desk on Friday, September 4, 2026. Its sources were placed by the desk, never implied. Open each step to go deeper; every hash says what it covers.

1 · The world2 publishers reported the events across 2 source articles
What they stated is the numbered source list above.
Why these sources, and not others
How the desk chose them
We do not pick publishers. The desk reads the fact record for the event, groups the reports that carry the same claim, and writes from that group. Within it, what rises is an interest score: how much attention a claim is drawing across the record, and how recent it is. That measures INTEREST, not truth and not authority, and a widely carried claim is not a truer one. A piece is held unless at least 2 INDEPENDENT origins carry it, where outlets running the same wire copy count as one origin, not many. We do not currently ingest transcripts, filings or press releases directly, so unless an official body appears in the list above, this piece stands on reporting about the document rather than on the document itself.
Where they publish from
1 of the 2 publishers publishes from Peru. We do not know which country the other one publishes from. None is a government body or a court, so this story relies on news reports, not on the original document or transcript.
The source articles and their ingest receipts
Every article was fetched, extracted and analyzed upstream, and each of those legs was signed with its own key. This opens the record's own receipts for them.
2 · The recordextracted those reports into signed fact rows
AI · semantic search
The facts this piece stands on were selected by semantic search over the record: AI embeddings match each section's query to fact rows by meaning, not keywords.
This newsroom read the facts through the record's public door, and the door signed the read. The read receipt was not captured for this early revision.
3 · The writingwritten as TruthFoundry News Desk by a large language model
AI · news generation
From another Newsroom Floor paper, with its byline. Each piece opens on the paper that wrote it.
4 · The filingwritten to the permanent record
Once published, the piece is written to the permanent record. Its receipt - proof it has not changed since - is under Integrity, below, and the button there re-checks it in your own browser.
Integrity
Content hash (SHA-256)53731f83436f54ebc080d127fa64d1b2bafe62272b99c67e2fb861d88763b749
Hash basisheadline + dek + prose + the canonical citations JSON, exactly as filed
Receiptthis revision predates receipt-keeping; the filed row lives on the record
Machine readablethe full proof, JSON
Verify

A signature proves who filed this and that it has not changed since. It never makes a claim true.