# FACTWASH: Catching AI Rewrites That Wash Hearsay into Fact

arXiv: https://arxiv.org/abs/2608.03372 · HTML: https://mlsanity.com/paper/2608.03372/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 19 references that are in the map.
- First posted on arXiv 4 Aug 2026.

## Builds on (most cited first, 5 of 5)

- [MemGPT: Towards LLMs as Operating Systems](https://mlsanity.com/paper/2310.08560/) (2023)
- [Understanding Factuality in Abstractive Summarization with FRANK: A Benchmark for Factuality Metrics](https://mlsanity.com/paper/2104.13346/) (2021)
- [Understanding Factual Errors in Summarization: Errors, Summarizers, Datasets, Error Detectors](https://mlsanity.com/paper/2205.12854/) (2022)
- [Reclaim Evaluation: A Lossy Memory Is Worse Than an Empty One](https://mlsanity.com/paper/2606.25449/) (2026)
- [Manufactured Confidence: How Memory Consolidation Turns Hearsay into Confident Facts](https://mlsanity.com/paper/2606.29279/) (2026)

## Built on by (most cited first, 1 of 1)

- [Explicit, Not Longer: What Makes Epistemic Stance Survive Memory Compression](https://mlsanity.com/paper/2608.06953/) (2026)

## Abstract (arXiv)

AI systems rewrite information constantly: conversations become stored memories, documents become answers. The rewrite can keep a claim while washing away what made it checkable, who said it, how sure they were, when it held. We call that failure factwashing, and release factwash, an open-source write-time gate that catches it deterministically, with named flags and evidence rather than an LLM judge. Building it answers a practical question: when does a cheap check suffice, and when do you need a model? What decides is whether the property has a bounded surface-cue inventory. Explicit negation cues are close to enumerable, so a word list finishes and transfers, reaching 0.91 F1 on untuned text. Hedging and attribution have open-ended realizations, so vocabulary plateaus near half recall, and a one-question LLM witness recovers +17 and +15 points of cue-detection recall at equal precision. Deployed, that witness may only lower a verdict, so it buys precision rather than coverage. We measure cue detection on 105,596 independently annotated sentences. A blind-labelled corpus of memory writes then locates the failure: 55% of bad writes in conversational hearsay, 7% in business email (p < 0.001), so the first deployment question is not which detector to use but whether the failure occurs at all. On unmodified mem0 2.0.7, the gate flags 5 of 8 hedged-hearsay writes.