HumanFlow is not one of the two tools on this page. We sell a rewriter and run a detector of our own, so weigh this accordingly — our methodology sets out how we source what appears below and what we refuse to claim.
One of these is not an AI detector. SafeAssign matches text against a database and reports overlap; it has no AI detection, and Anthology decided against adding it.
Last reviewed 16 August 2026 · The HumanFlow team
Students compare these two because both turn up inside a university's systems and both produce a percentage, so it is reasonable to assume they are doing the same job. They are not. The confusion is common enough that an entire genre of misleading content exists to serve it, quoting SafeAssign AI-detection accuracy figures that cannot exist.
It is also a real procurement question. An institution running Blackboard has SafeAssign already, and the question of whether that is enough for AI has a straightforward answer that Anthology has published.
This is the only comparison in the family where the useful outcome is usually a correction rather than a decision. Most people arriving here have been told, or have assumed, that a SafeAssign percentage says something about AI use. It does not, and establishing that is worth more than any recommendation — which is why the page leads with the distinction and treats the procurement question as secondary.
No prices. Neither of these is a tool most readers choose on cost, and for one of them there is no public price at all — detector pricing, where it could be captured, is on our detector pricing page.
| Dimension | SafeAssign | Copyleaks |
|---|---|---|
| Who it is sold to | Institutions, bundled inside Blackboard. Owned by Anthology, which evaluated adding AI detection and decided against it. | Institutions and individuals both — one of the few detectors genuinely sold to universities rather than only to people. Education is priced on full-time student count, with integrations for Canvas, Moodle, D2L Brightspace, Blackboard, Schoology, Sakai and Edsby. |
| How you get at it | No AI detector to access. It is a text-matching tool and reports overlap only. | Paid. Personal is $16.99 a month, or $13.99 a month billed annually; one credit covers up to 250 words. |
| What it needs | Not applicable — it matches text against a database rather than judging prose. | Not published as a word floor. Its highest sensitivity setting is described as designed to flag text put through a humanizer or spinner. |
| What independent testing found | Not applicable as an AI detector. Anthology's own 2023 beta across 65 client institutions and 1,000+ texts found 80% of respondents said the detection tool tested was at best only “sometimes” correct, and Anthology concluded AI detection was “not currently fit for purpose in education”. | Best of the seven tools in Perkins et al. (2024) — and it still missed 39% of the AI cases and produced the highest false-accusation rate in that set. Its accuracy fell from 73.9% to 58.7% once simple adversarial techniques were applied. |
| What the vendor claims | None for AI detection, because it makes none. Any page quoting a “SafeAssign AI detection accuracy” percentage has invented it. | 99.97% accuracy and a 0.026% false-positive rate on its V10 model. |
| What the vendor says about its limits | Anthology's guidance to institutions recommends authentic assessment design instead of detection. | No published statement we could find. |
| What happens to your text | The institution's copy goes to its Institutional Database by default. A separate Global Reference Database is voluntary, and Anthology's student guidance states that submitting to it means agreeing not to delete papers in future. Exclusions must be set before students submit and cannot be applied retroactively. | Paid plans include the Shared Data Hub, which Copyleaks describes as a library of user-submitted documents that scans are compared against. Its privacy policy states it uses this information to train its models, with opt-out available to direct customers by contacting support. |
Every line above is summarised from our own examination of each tool, where the studies and vendor documents behind it are quoted and linked: SafeAssign and Copyleaks.
SafeAssign is a text-matching tool. It compares a submission against a database of academic papers, publications, articles and internet sources and reports overlap. Generative AI produces original text that matches nothing, so there is nothing for SafeAssign to match it against. A SafeAssign score means “this wording appears somewhere else”, not “a machine wrote this” — and a low score is not a clearance either.
It can still flag AI-written text, for an unrelated reason: if the model reproduced phrasing that exists elsewhere online, that overlap matches like any other. Reading that as AI detection is the specific error this page exists to prevent.
Copyleaks does both. It is a plagiarism checker with a genuine AI detector attached, sold to institutions on full-time student count with integrations for Canvas, Moodle, D2L Brightspace, Blackboard, Schoology, Sakai and Edsby — including Blackboard, so the two can run alongside each other rather than instead of each other.
The most useful part of this comparison is why the gap exists. Anthology, which owns Blackboard, ran a beta with market-leading AI detection tools across May and June 2023 — 65 client institutions, more than 1,000 texts, some authentic and some AI-generated — and concluded, in its own white paper, that “AI detection is not currently fit for purpose in education”. 80% of respondents said the tool was at best only sometimes correct. One participant generated six ChatGPT texts and reported the detector misidentified all six as human. Its guidance recommends authentic assessment design instead.
The interface is doing real damage here and it is worth being concrete about the mechanism. Both tools return a percentage in a similar-looking report at the same point in a student's week, and a percentage invites the same reading regardless of what produced it. A SafeAssign score of 34% and a Copyleaks AI score of 34% are not comparable quantities, are not measuring the same property, and do not support the same inference — but they look identical on a page.
There is no SafeAssign AI accuracy figure, because there is no SafeAssign AI detector. If you find a page quoting one, that tells you what the page is worth. Anthology's beta is a survey of what participating institutions concluded rather than a controlled measurement, and it should be read as that — but it is a vendor with every commercial reason to ship a detector explaining why it did not.
Copyleaks has been independently tested. In Perkins et al. (2024) it was the best of seven tools — and it still failed to identify 39% of the AI-generated cases and produced the highest false-accusation rate in that set. Under simple adversarial conditions its accuracy fell from 73.9% to 58.7%. Its own published figures are 99.97% accuracy and a 0.026% false-positive rate on its V10 model.
That gap between the vendor's number and the independent one is the most informative thing on this page. Both are real measurements of different things, under different conditions, on different text. Anthology's conclusion and Copyleaks's marketing are not actually in contradiction: a tool can be the best available and still be unfit to decide a misconduct case alone.
Anthology's decision is the most underused piece of evidence in this entire category, and its value lies in who made it. A learning-platform vendor with an obvious commercial incentive to ship detection, and access to 65 client institutions to test it with, concluded the technology was not fit for the purpose. That is a supplier's finding against its own commercial interest, which is the rarest and most credible form of evidence available anywhere in this market.
SafeAssign predates this entire category. It is a text-matching tool built into Blackboard for plagiarism checking, and it has been doing that job largely unchanged since long before a language model could write an essay. It was never redesigned for AI because it was never asked to detect AI — a fact that gets lost because it sits in the same interface, produces a percentage, and appears at the same moment in a student's week.
Anthology, which owns Blackboard, did consider adding detection and published its reasoning for not doing so. Across May and June 2023 it ran a beta with market-leading AI detection tools across 65 client institutions and more than 1,000 texts. Its white paper concluded that AI detection is “not currently fit for purpose in education”, and 80% of respondents said the tool tested was at best only sometimes correct. One participant generated six ChatGPT texts and reported the detector misidentified all six as human.
Copyleaks made the opposite call, and it is worth seeing them as two answers to one question rather than as a good product and a bad one. A company whose business is content integrity shipped detection with sensitivity settings and a documented position on humanized text; a company whose business is a learning platform tested it, decided the error rate was unacceptable for the population it would be applied to, and recommended assessment design instead.
If the number came from SafeAssign, it is not an AI finding and cannot be turned into one. SafeAssign reports overlap between your text and a database of papers, publications, articles and internet sources. Generative AI produces original text that matches nothing, so a high SafeAssign score means your wording appears somewhere else — a quotation, a common phrase, a properly cited passage, a template — and a low score is not a clearance either.
This confusion is common enough to be worth naming precisely when it happens. If an instructor has described a SafeAssign similarity percentage as an AI score, the correction is factual rather than adversarial: the tool does not have that capability, and Anthology has published why it decided against adding it.
If the number came from Copyleaks, it is a genuine AI detection result and the response is different. Ask which sensitivity setting was used, because the highest one is explicitly designed to flag humanized and spun text and accepts more false positives to do it. Then produce your drafting record, which is the evidence that speaks to how the document was actually made.
In both cases, Anthology's own conclusion is a legitimate thing to put in front of a committee. A platform vendor with every commercial reason to ship a detector tested one across 65 institutions and declined. That is not a student's opinion about detection being unreliable; it is a supplier's.
There is no SafeAssign AI accuracy figure and there cannot be one, because there is no SafeAssign AI detector. Any page quoting a percentage for it has invented the number, and its presence is a reliable signal about the rest of that page.
Anthology's beta is a survey of what participating institutions concluded, not a controlled measurement. It reports how accurate 65 institutions found the tools they tested; it does not name them, and it does not produce a rate anyone can cite as a benchmark.
We could not establish which detector Copyleaks was tested against, or whether it was among the tools in Anthology's beta at all. The white paper describes “market-leading AI detection tools” without naming them.
Copyleaks pricing was read on 15 August 2026 and could not be re-verified on 17 August: the page now returns a Cloudflare challenge to automated requests. The figures on this site were read from the page and stand, but they need a browser read to be confirmed current.
We could not establish how many institutions run a third-party detector alongside SafeAssign. That number would tell a student more than anything else on this page — because 'Blackboard has no AI detection' does not mean 'nothing is checking your work', and the gap between those two statements is where most of the confusion in this area actually lives.
Pick SafeAssign if the question is plagiarism rather than AI. For matching text against sources, SafeAssign does the job it was built for, and it is already licensed wherever Blackboard is.
Pick Copyleaks if you need AI detection at all, because SafeAssign offers none. Read what the Shared Data Hub does with submissions first, and read Anthology's reasoning before treating any AI score as decisive.
A detector reports how statistically machine-typical a piece of prose reads. It has no access to how the text was produced, so it cannot establish authorship in either direction — a flag is not evidence of AI use, and a clean result is not a clearance. If you have been accused on the strength of one, drafting history is what actually answers it. And if your institution requires you to disclose AI assistance, disclose it — nothing on this page changes that obligation.