AI detector census 2027: how many detectors the index holds, how many are gone, and how many have ever been tested
72 AI detectors have a card in the index at version 2026.11.13, and 12 of them are withdrawn, closing, unreachable or redirected, by the index data file read on 10 November 2026 (the AI detector index); 35 of the 72 have at least one study record, 23 have a false positive rate printed by a study rather than the vendor, 8 are open source, and 4 register rows planned for the year were researched and held unpublished. Every count on this page was computed from the data file and the study list on that date and is stated with its rule, so a reader a year from now can recompute it; the how many AI detectors exist question keeps the running count between censuses, and this is the second annual piece after the 2026 false positive evidence review.
Checked 10 November 2026 for the 2027 census year, the year the twelve-month content plan closes; the next scheduled recheck of every card is 8 December 2026.
This is the annual census. It counts what the index holds, not what exists: no register of AI detectors exists anywhere, tools open and close without notice, and the largest study in the index tested 16 (Walters 2023 record, DOI 10.1515/opis-2022-0158, read 8 November 2026). The counts are grouped by what they answer, one table per section: how many and how many are gone, what media they take, how many have been tested and with what result, how many are open or held, what the study index itself holds, and when the detectors say they launched. This page sits in the updates hub beside the Q3 2027 index update and the Q3 2027 studies digest, which cover the last quarter's additions.
| Count | Value on 10 November 2026 | Rule | Where it is kept |
|---|---|---|---|
| Cards in the index | 72 | One card per detector read on the same fields | Detectors hub |
| Withdrawn, discontinued or closing | 8 | The card's status prose says the maker withdrew, discontinued or is closing the tool: Writer, Wordtune, Check For AI, Writefull, Passed.ai, GPT Radar, Grammica, AI Writing Check | Shut-down question |
| Unreachable, no statement | 1 | Deepware: scanner host unreachable and the Play listing 404 on 9 and 10 November 2026 (play.google.com) | Deepware card |
| Redirected to another product | 2 | BrandWell and ContentDetector.ai, both 301 to a Moxby marketplace Mod | Shut-down question |
| AI verdict withdrawn from a live product | 1 | Codeleaks: the AI source-code verdict removed from the submit endpoints on 16 December 2025 | Codeleaks card |
| Withdrawn with a record and no card | 1 | OpenAI's AI Text Classifier, withdrawn 20 July 2023 (openai.com, read 8 November 2026), carried as a study record | OpenAI classifier record |
| Live as read | 60 | 72 minus the 12 rows above; two of the 60, the GPT-2 Output Detector and DetectGPT, have a paused or vanished hosted demo and code that still runs | Launch year ranking |
Every row counted from the data file on 10 November 2026; the twelve are 12 of 72 cards, one in six.
What media do the 72 detectors take?
Text is the modality of 58 of the 72 cards, image of 20, video of 15, voice of 15, music of 5 and code of 4, by the modalities field as each vendor states it and as the content type matrix lays it out on 10 November 2026; two cards, Winston AI and Compilatio, also state handwriting through OCR (handwriting and OCR view). The counts sum to more than 72 because a card can state several modalities: Hive and AI or Not state five each. The two annual counts that matter beside modality are stated fields, not quality: 39 cards state an available API (API view), 31 state a language count (languages ranking), 14 name an LMS (LMS coverage ranking), 21 state a retention period in days (data retention ranking), 23 state a check with no account (no-account view), 17 a bulk route, 12 a named compliance standard, 19 a deployment option and, from the field written on 10 November 2026, 20 a seat count or single sign-on (team seats and SSO view).
| Modality stated | Cards | Study records on that modality | Cards with a record on that modality |
|---|---|---|---|
| Text | 58 | 27 | 31 |
| Image | 20 | 2 | 4 |
| Video | 15 | 2 | 0 |
| Voice | 15 | 4 | 0 |
| Music | 5 | 0 | 0 |
| Code | 4 | 0 | 0 |
Cards from the modalities field and records from the study list on 10 November 2026; the voice and video records are challenges that score no indexed card (studies by content type ranking).
How many of the 72 have ever been tested?
35 of the 72 cards carry at least one study record and 37 carry none, by the study count on each card on 10 November 2026 (independently tested ranking). The distribution is steep: 15 cards have exactly one record, 7 have two, and three cards hold 38 of the 112 record attachments in the index between them, GPTZero with 16, the GPT-2 Output Detector with 13 and Copyleaks with 9. A false positive rate is on 26 cards, 23 of them from a study and 3 from the vendor alone (Proofademic, Codequiry and Undetectable AI), by the published false positive field (false positive rate ranking); a false negative rate is on 20, an adversarial robustness figure on 15 and a claim-versus-evidence gap on 12. On the other side of the ledger, 54 cards quote a percentage accuracy figure from the vendor (claim vs evidence gap ranking).
| Records on the card | Cards | Examples |
|---|---|---|
| 0 | 37 | Every voice, video and music card except the four image cards scored by Ha and Passananti 2024 and RAID image 2025 |
| 1 | 15 | Scribbr, Compilatio, PlagiarismCheck, Fast-DetectGPT, SEO.ai |
| 2 | 7 | QuillBot, Winston AI, Hive, GPTKit, RADAR |
| 3 to 5 | 8 | Sapling 3, Pangram 4, GLTR 4, Turnitin 5, Writer 5 |
| 7 to 9 | 3 | ZeroGPT 7, Originality.ai 7, Copyleaks 9 |
| 13 and 16 | 2 | GPT-2 Output Detector 13, GPTZero 16 |
Counted from the study count field on 10 November 2026; a record counts once per card however many figures it prints.
How many are open source, and how many rows were held?
Eight cards are open source, by the open-source view read 10 November 2026: GLTR, the GPT-2 Output Detector, DetectGPT, SynthID, Fast-DetectGPT, Binoculars, RADAR and Ghostbuster, six of them with a study record (the GPT-2 Output Detector 13, GLTR and DetectGPT 4 each, Binoculars and RADAR 2, Fast-DetectGPT 1) and two with none (SynthID, Ghostbuster). Four register rows of the 189 cleared for the year were researched and held: two cards and two study records, each for a reason the index states rather than infers.
| Held row | Register ID reserved | Reason stated | Decision owner |
|---|---|---|---|
| Imagga (ADI-144) | AIDI-D-0052 | The vendor publishes no AI-generated image detector on any live page, documentation section or archive capture read 10 November 2026 (imagga.com, docs.imagga.com) | Kevin: strike or keep |
| Optic (ADI-153) | None | Optic was AI or Not's prior name; the AI or Not card and the Ha and Passananti 2024 record already carry it, so a second card would duplicate one product | Kevin: strike or keep |
| Dalalah 2023 (ADI-078) | AIDI-S-0020 | Closed access; no primary copy reachable on 9 November 2026, and the index does not summarise a paper from a write-up | Kevin: hold or source |
| Jiang 2024 (ADI-084) | AIDI-S-0022 | Closed access; the same rule | Kevin: hold or source |
Rows from the topic register on 10 November 2026; the Q2 2027 index update recorded the Imagga hold and the Q2 2027 studies digest the two papers.
What does the study index hold, by year and kind?
35 records are in the studies index on 10 November 2026, thirty-five in words, and 2023 is the year of 17 of them, by the study year in each record's front matter. By kind, 21 are independent evaluations, 6 are challenges (ASVspoof 2021 and 5, ADD 2023, SAFE 2025, DFDC 2020, FaceForensics++ 2019), 5 are vendor or lab reports, and one each is a human-rater study (Fleckenstein 2024), a data insight (Epoch AI 2026) and an institutional statement (Vanderbilt 2023), by the kind label each record carries on the hub. The six challenges and the human-rater study score no indexed card, which is why 35 records feed only 35 cards.
| Study year | Records | Kinds in that year |
|---|---|---|
| 2019 | 1 | Challenge (FaceForensics++) |
| 2020 | 1 | Challenge (DFDC) |
| 2023 | 17 | 12 independent, 2 vendor or lab, 2 challenges, 1 institutional statement |
| 2024 | 5 | 4 independent, 1 human-rater study |
| 2025 | 6 | 4 independent, 1 vendor or lab, 1 challenge |
| 2026 | 5 | 1 independent, 2 vendor or lab, 1 challenge, 1 data insight |
Years from each record's stated study year on 10 November 2026; ASVspoof 2021 carries 2023, the year of its published results.
When do the detectors say they launched?
29 of the 72 cards state a launch year on a vendor or author page, and 43 state none, by the launch year ranking read 10 November 2026; the index enters no year from an archive's first capture. Of the 29, 2023 is the launch year of 14, from GPTZero and Turnitin to SynthID, Fast-DetectGPT and RADAR; the earliest is Codequiry in 2017 and the latest Turnitin Clarity in 2025. Three of the 12 gone rows state a year, Wordtune, AI Writing Check and BrandWell, and all three launched in 2022 or 2023 and were gone by 2026.
| Launch year stated | Cards | Named |
|---|---|---|
| 2017 to 2019 | 4 | Codequiry, Sensity, GLTR, GPT-2 Output Detector |
| 2021 | 2 | Reality Defender, Deepfake-O-Meter |
| 2022 | 4 | Originality.ai, BrandWell, Illuminarty, FakeCatcher |
| 2023 | 14 | Pangram, GPTZero, Turnitin, AI or Not, SynthID, DetectGPT, Codeleaks, AI Detector Pro, Fast-DetectGPT, Wordtune, ElevenLabs classifier, AI Writing Check, RADAR, Ghostbuster |
| 2024 | 4 | Pindrop Pulse, Binoculars, Hiya, McAfee |
| 2025 | 1 | Turnitin Clarity |
| None stated | 43 | The remaining cards |
Years as the cards state them, from the launch year field on 10 November 2026.
No detector score should be the only evidence in a misconduct, hiring or publishing decision, and a census of 72 tools with 35 tested is a count of what has been written down, not of what works.
Questions about the 2027 census
How many AI detectors are there in 2027?
72 have a card in the index at version 2026.11.13 on 10 November 2026, 12 of which are withdrawn, closing, unreachable or redirected; no published census counts more, and the largest study tested 16 (Walters 2023 record).
How many AI detectors have been independently tested?
35 of 72 carry at least one study record on 10 November 2026, and 23 carry a false positive rate printed by a study; GPTZero has the most records at 16 (independently tested ranking).
How many AI detectors have shut down?
Eight cards record a withdrawal, discontinuation or closing (Writer, Wordtune, Check For AI, Writefull, Passed.ai, GPT Radar, Grammica, AI Writing Check), one is unreachable (Deepware), two redirect (BrandWell, ContentDetector.ai), one withdrew its AI verdict (Codeleaks), and OpenAI's classifier is a record with no card, by the pages and captures the shut-down question dates.
How many AI detectors are open source?
Eight on 10 November 2026: GLTR, GPT-2 Output Detector, DetectGPT, SynthID, Fast-DetectGPT, Binoculars, RADAR and Ghostbuster (open-source view).
How many planned pages were not published?
Four register rows of 189: the Imagga and Optic cards and the Dalalah 2023 and Jiang 2024 records, each held for the reason in the table above and awaiting a decision; the other 185 rows are written and the index holds 237 pages.
Sources
- The index data file, version 2026.11.13 (read 10 November 2026)
- The topic register, 191 rows (read 10 November 2026)
- The study list and the kind, modality and tested labels on the studies hub (read 10 November 2026)
- Walters 2023: https://www.degruyterbrill.com/document/doi/10.1515/opis-2022-0158/pdf (read 8 November 2026)
- OpenAI classifier statement: https://openai.com/index/new-ai-classifier-for-indicating-ai-written-text/ (read 8 November 2026)
- Deepware Google Play listing (404): https://play.google.com/store/apps/details?id=com.zemana.deepware (read 9 and 10 November 2026)
- Imagga (held): https://imagga.com/ and https://docs.imagga.com/ (read 10 November 2026)
Revision log: 2026-11-10: census created at index version 2026.11.13, the second annual piece (twelfth wave).
Related pages: Index update, Q3 2027: twenty-one cards added, two fields added, two rows held, three log lines, Studies digest, Q3 2027: five records added, one headline figure and base each, two closed-access papers still held, AI detectors ranked by languages supported, How accurate are AI detectors? Every study, indexed and Index updates: dated reviews of the AI detector data.