# Knowledge-cutoff disclaimer Part of [The AI Tells Index](https://feedsquad.com/ai-tells). A tell signals low effort. It does not identify an author. Skilled writers produce every shape listed here, some of them daily, and automated detectors misread those writers for it at rates measured above 60 percent on non-native English prose. Nothing in this index proves that a machine wrote anything. Read an entry as one piece of evidence to weigh against the false-positive notes printed beside it. ## Facts - Id: knowledge-cutoff-disclaimer - Category: Model register (model-register) - Subcategory: assistant-leak - Also known as: As of my last training update, as of my knowledge cutoff - Status: Active. Currently signals low-effort writing. - Severity: high - Evidence grade: corroborated (named independently by multiple credible secondary sources) - Languages: en - Added: 2026-08-15 - Updated: 2026-08-17 - Page: https://feedsquad.com/ai-tells/knowledge-cutoff-disclaimer ## Description The text dates itself to a training window. As of my last training update, followed by a fact that may be two years old. Inside a chat window the disclaimer is honest. On a page with a byline and a publication date it contradicts both. The sibling entry for an undisclosed stale claim covers the fact. This one covers the sentence admitting it. ## Why it reads as machine-written Nobody has a training update. Wikipedia names knowledge-cutoff disclaimers among the examples of communication intended for the user under speedy-deletion criterion G15, and the tellsign list files the same phrasings as wording no person would write about their own knowledge. Two failures ship in one sentence: a fact that was never checked, and an admission that it was not. ## Detection Type: deterministic (a regular expression, run as written) Scope: sentence ECMAScript regular expression. The build validator compiles it and tests it against this entry's own specimens. ```regex \b(?:as of|up to|until)\s+my\s+(?:last\s+)?(?:training\s+(?:update|data|cut-?off)|knowledge\s+(?:cut-?off|update))\b|\bmy\s+training\s+data\s+(?:only\s+)?(?:goes|extends|runs)\s+(?:up\s+)?(?:to|through)\b ``` Flags: i ## Examples Constructed specimens. Written for this index. Never quoted from anyone's posts. ### Specimen 1 Before, exhibiting the tell: > As of my last training update, the company ran a handful of offices and employed a few dozen people. It has grown steadily since then, in line with the wider market, and nothing about the picture appears to have changed in any material way. Companies at that stage rarely change course, and when they do it is generally announced rather than discovered. After, repaired: > The company ran two offices in the Nordics when we checked the Finnish trade register on 4 June 2026: Helsinki since 2019 and Oslo since 2024. Headcount was 41. Note: The repair swaps the training window for a source, a date and a number a reader can check. Figures in this repair are invented for the specimen. ### Specimen 2 Before, exhibiting the tell: > My training data only goes up to early 2024, so treat the pricing table below as indicative rather than current. Prices in this category move slowly, so the figures should still be close enough for most readers, and where something looks off the vendor sites will have the current number. After, repaired: > We last checked every price on this table on 12 August 2026, against the vendors public pricing pages. Two of the six had changed since June, both upward. Note: A hedge that admits nothing becomes a dated verification note. Figures in this repair are invented for the specimen. ## False positives Who legitimately writes this way. People who write about models quote the phrase constantly, and this entry is one of them. Product documentation for an assistant contains it in every release note. Published transcripts and retyped screenshots carry it legitimately, and so does any post arguing about what a model can know. Exclude quoted and transcribed spans first. What remains is a first-person training disclaimer inside copy that carries a human byline and a publication date. That is a contradiction on the page, not a finding about the author, and the repair is a checked date, not a deletion. ## Model attribution Common to assistant products across vendors, and the wording shifts with the system prompt, not with the weights. Some products now answer date questions from a tool call instead, which is why the sentence is seen less in 2026 output than in 2023 output. ## Platform notes - wikipedia: Named in criterion G15 as one of the examples that justify speedy deletion, in the same list as collaborative address and the self-naming sentence. ## Sources 1. Wikipedia: Criteria for speedy deletion, criterion G15 https://en.wikipedia.org/wiki/Wikipedia:Criteria_for_speedy_deletion (tier: community; accessed 2026-08-14) 2. tellsign word and phrase lists https://github.com/ctkrug/tellsign (tier: community; accessed 2026-08-14) ## Status history Ids are permanent. A retired tell keeps its id and its page. - 2026-08-15, Active: Active on Wikipedia criterion G15, which names cutoff disclaimers among the examples of communication intended for the user, and on the tellsign phrase list, which carries the same wordings independently. Pattern verified on 2026-08-15 against both specimens and against three ordinary sentences, including one about training records that end in 2019. ## License CC BY 4.0. https://creativecommons.org/licenses/by/4.0/ Attribution: The AI Tells Index, feedsquad.com/ai-tells Reuse the data, including commercially. Keep the attribution line. --- Dataset version 1.0.0. Schema version 1. Part of [FeedSquad](https://feedsquad.com). Built by [Herman Foundry](https://hermanfoundry.com) from Levi, Finnish Lapland.