Abstract
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.
Explore similar work
Dec 29, 2025cs.HC
Fact-checking systems must be scalable and epistemically trustworthy. We introduce Althea, a retrieval-augmented system for user-driven claim evaluation that matches standard pipelines on AVeriTeC while improving supported/refuted discrimination. A longitudinal survey experiment (N=961) treats a ten-day follow-up as a fading test: after modeling a verification procedure, we remove the system and ask whether users reproduce it unaided, testing metalearning rather than one-time accuracy. We compare two AI-assisted treatments, Exploratory (guided reasoning) and Summary (synthesized verdicts), against two baselines, unrelated news and Self-search. The treatments yield the strongest immediate accuracy and confidence gains but do not survive the fading test: on unseen claims they perform no better than news, while Self-search, with no procedure to fade, retains a large advantage. This reveals a factchecking-metalearning tradeoff: conditions that most improve immediate accuracy are least likely to produce metalearning, cautioning against treating AI-delivered verdicts as a source of durable literacy gains.
Svetlana Churina, Kokil Jaidka, Anab Maulana Barik +5
Jun 28, 2026cs.CR
LLM agents carry conclusions across steps and sessions in compressed memory, and memory products (e.g., mem0, LangMem) rewrite conversation into stored "facts" that later steps trust. We show this rewriting manufactures confidence: across our constructed agent settings, a casual, hedged remark becomes a confident, dated assertion the agent then obeys like a verified fact, granting every above-clearance request it faces. No attacker is needed: a role that was true once and never corrected is stored as a flat fact and acted on like a deliberate injection. We then isolate what the agent responds to. It is not the source: attributed, unattributed, and even forged "system of record" claims all grant alike. It is the confidence of the phrasing. A hedge is discounted, a flat assertion is obeyed, and this holds with no special keyword. Not all hedges are equal, though: the evidential register is the least-discounted, with "reportedly" obeyed like a flat assertion on most models. The obvious fixes fail. A passive "unverified" tag is ignored, and an active "do not trust this" instruction escalates even correct memory, so it is safe only by refusing to decide. The real fix lives in the store: keep the tentative phrasing rather than upgrade it. But that is hygiene, not a defense against an attacker who can simply write a confident lie. The deployable lesson is narrower and constructive: a single load-bearing memory is the hazard, and one redundant source restores correct decisions. We release the harness and demonstrations.
Alex Kwon
Jun 23, 2026cs.CL
Fact-checking systems built on LLMs achieve high verdict accuracy on standard benchmarks, yet routinely output Supports labels whose cited evidence does not license the claim. Structured decomposition is the natural way to inspect those warrants, but rigid extraction protocols strip the full-claim context that facets need. We introduce SIFT -- claim-conditioned re-scoring of extracted evidence spans against the full claim -- paired with WSP (Warranted Supports Proportion), an automatic NLI check that the cited warrant entails the claim. We evaluate on FEVER, SciFact, 5PILS, and DP across four open-source backbones. SIFT recovers accuracy on cells where naive decomposition costs up to 27.6 points, while raising WSP above direct prompting; WSP itself calibrates against human gold evidence at AUC 0.92 and precision 0.98.
Arka Ujjal Dey, John Collomosse