SafeAssign AI Checker: Does Blackboard Detect AI Writing?

Who Is This SafeAssign Guide For?

This guide answers one question for four readers: students using Blackboard who saw a SafeAssign score, non-native English (ESL) writers worrying about being misclassified as AI authors, professors and instructors reading a Blackboard detection report, and academic administrators weighing whether to license an AI-detection plugin at all. What it will not do is help anyone hand in work they did not write.

A clarification about HumanizeMyAI. HumanizeMyAI does not integrate with SafeAssign or Blackboard. We are an independent humanizer plus AI-probability checker. Our free AI probability check is a pre-submission self-verification tool, separate from any institutional Blackboard report. We received no compensation from Anthology, Copyleaks, Turnitin, or any tool mentioned.

Does SafeAssign Detect AI Writing?

SafeAssign does not detect AI writing, so it cannot catch ChatGPT. Anthology, the company that owns Blackboard and SafeAssign, has never shipped a SafeAssign feature that reads writing style. Its own product documentation describes the service as plagiarism similarity matching against databases and the open web, with no style analysis anywhere in the pipeline.

When a Blackboard instructor tells a student “your work was flagged as AI,” the flag came from a different tool. It came from a third-party plugin the school installed on top of Blackboard, most often Copyleaks or Turnitin, or from a manual instructor judgment made outside the platform entirely. Which of those it was is knowable, and the next two sections show how to tell.

What Does SafeAssign Actually Check?

SafeAssign is a source-matching plagiarism tool. It breaks the submission into sentence-level passages and compares each passage against four corpora: the internal SafeAssign reference database, ProQuest ABI/Inform, the institutional database of previously submitted assignments, and the open internet.

That is the entire mechanism. There is no language model in the pipeline. There is no perplexity calculation. There is no burstiness analysis. SafeAssign cannot tell whether a passage was written by a human or by ChatGPT, because it does not analyze writing style at all. It only checks whether the passage already exists somewhere else. For a broader plagiarism tool comparison, see our plagiarism checker hub.

Does the SafeAssign Originality Report Detect AI?

The SafeAssign Originality Report does not detect AI. The name causes the confusion: it is simply the current official label for SafeAssign’s output, not a second, newer product sitting next to it. Anthology’s instructor documentation describes the report as information about matches found between a submitted paper and existing sources, and never mentions AI, language models, or writing style.

The word “originality” is doing a lot of unintended work in 2026. To SafeAssign it means “not copied from a source we have on file.” To a worried student in 2026 it sounds like “not written by a machine.” Those are different questions, and the report only answers the first one. If you are looking at a SafeAssign Originality Report, there is no AI verdict on it to find, no matter how carefully you read it.

What Does Your SafeAssign Score Mean?

Your SafeAssign score reports how much of your text matched a source on file, not how likely it is that AI wrote it. Anthology publishes three risk bands for that matching percentage, and the table below carries its reading of each.

Two things worth knowing about that number. First, the bands are guidance for a human reader, not a verdict: a heavily quoted literature review can land in Medium and still be completely legitimate work, which is why the documentation tells instructors to open the report and check whether matches are properly attributed rather than judge by the percentage alone. Second, the per-sentence numbers mean something different from the overall score. A sentence marked 90% is an estimate that the two phrases carry the same meaning, not a claim that 90% of your sentence was stolen.

What AI Detector Does Blackboard Use?

Blackboard Learn ships no AI-writing detector of its own, so there is no default answer. The tool that flagged you is whichever one your institution chose to install and pay for, and two of them show up most often.

Copyleaks builds and sells its own Blackboard plugin, delivered as an LTI integration with separate setup paths for instructors and students. Turnitin is licensed the same way, and brings the AI Writing Indicator described on our Turnitin AI checker guide. Both are purchases your school makes on top of Blackboard, not features that arrive with it.

Anthology’s own position is worth knowing, because it cuts against what most pages on this topic imply. Its administrator guidance on AI plagiarism names no detector at all, recommends nobody’s product, and spends its length arguing for assessment design instead: tasks built around critical thinking, personal perspective and self-reflection, peer assessment, and instructors becoming familiar enough with a student’s writing to notice when it changes. A student who reads that page and expects to find the official Blackboard AI detector named will not find one, because Anthology has not put one there.

You may also read that Copyleaks became Anthology’s official AI-detection partner in 2023. We published that claim ourselves and can no longer support it: no Anthology or Copyleaks announcement we can find states it. The dated 2023 partnership on record is between Copyleaks and Moodle, announced April 21, 2023, and Copyleaks was named an industry-recommended AI and text-matching partner for Canvas on June 25, 2026. Neither of those is Blackboard. What is verifiable is narrower and more useful: Copyleaks sells a Blackboard integration, your school may or may not have bought it, and the only way to be certain is to ask.

How Can You Tell If Blackboard Is Checking Your Work for AI?

Blackboard leaves you in one of three states, and what the report is branded tells you which. If only SafeAssign appears, nothing on your submission is being read for AI at all. A second score beside it means a plugin your school bought is doing that work.

Ask your institution's academic integrity office which configuration is active. The Blackboard submission interface does not disclose this.

What Should You Do If Blackboard Flags Your Work as AI?

Answer the AI flag separately from the similarity score, because two unrelated systems wrote those numbers onto one screen and only one of them was asked about AI. The combination that catches people out is a fine SafeAssign score, say 12%, sitting beside a separate “AI Content” flag. Nothing has gone wrong with SafeAssign.

Argue with the flag on its own terms, then. Ask your instructor for the full detector breakdown rather than the headline number, since these reports list which sentences were flagged and how confidently, and a flag concentrated in your methods section reads very differently from one spread evenly through the paper. Copyleaks refreshed its AI Content Detector reporting in early 2026, and our pre-submission Copyleaks guide walks through how it scores text.

Then run the same paragraphs through a second, unrelated detector such as ours. Where two independent classifiers disagree about the same paragraph, that disagreement is a fact about the classifiers, and it belongs in front of whoever is reviewing your case. No detector on the market has published accuracy above 97% on real student writing, which is a long way from the certainty a single flag tends to project.

What Is the SafeAssign Draft Box?

The SafeAssign draft box is the closest thing to a self-check the tool offers, and whether you get one is your instructor's call rather than yours. It appears when the instructor, building the assignment, ticks Exclude submissions from the Institutional and Global References Databases. Anthology’s student documentation is specific about what that changes: the submission is stored as a draft in their internal database for technical purposes, no other submission can be matched against that draft, and the draft can still generate an Originality Report so you can see your matching score.

In practice that means the assignment you are about to hand in cannot be flagged as matching your own earlier draft of it, which is the failure mode students actually fear. What it is not is a private tool you control. Turnitin sells Draft Coach, which a licensed student can open whenever they like. The SafeAssign equivalent only exists if your instructor built it into that specific assignment. If the option is not there, no setting on your side will produce it, and the request to make one available goes to the instructor.

This is also the honest answer to a question a lot of people arrive here with: there is no free public SafeAssign checker. SafeAssign runs inside a licensed Blackboard instance against a database that includes your own institution’s past submissions, and no outside site has access to either. Tools advertising a free SafeAssign check are running something else and calling it SafeAssign. If a draft box was not offered to you, the workable pre-submission checks are your own citation formatting and an independent AI probability check, which answer different questions than SafeAssign does but are at least yours to run.

Is SafeAssign a Plagiarism Checker or an AI Detector?

SafeAssign is a plagiarism checker, and only that. It cuts a submission into sentence-level passages and matches them against four corpora: its own reference database, ProQuest ABI/Inform, the archive of work previously submitted at your institution, and the open web. Style never enters that pipeline, so the tool produces a similarity figure and cannot produce an AI probability at all.

Any AI number you have been shown inside Blackboard therefore came from somewhere else: a plugin your school licensed on top of the platform, most often Copyleaks or Turnitin, or an instructor judgment made off-platform. That is why the two scores answer different questions, which the next section takes apart.

How Is a Similarity Score Different From an AI Score?

A SafeAssign similarity score and an AI probability score measure different things, and confusing them is the most expensive mistake on this page. Students see a 25% or 35% similarity figure and conclude that 25% or 35% of their writing was identified as AI. Nothing in that number is about authorship.

• Similarity (SafeAssign): Is this text reusing words from a source on file? (A textual-overlap question.)
• AI probability (the plugin, not SafeAssign): Was this text likely generated by a language model? (A statistical-style question.)

Plenty of legitimate, well-cited, fully human academic work returns SafeAssign similarity scores in the 20% to 40% range because academic writing quotes sources. The fix for an unexpectedly high SafeAssign score is to verify quotation marks and citation formatting, not to rewrite the text to sound less like AI.

Do False Positives Hit Honest and ESL Writers?

False positives reach honest writers by two separate routes. SafeAssign similarity scores run high on fully human work through common-phrase overlap and properly quoted sources, which is not an AI finding at all. For non-native English writers, the AI detector layered on top carries a measured bias, 61.3% on authentic TOEFL essays.

For non-native English (ESL) writers, the stakes get higher. Stanford researchers published a 2023 study in Patterns (Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. DOI 10.1016/j.patter.2023.100779). The Stanford team found that AI classifiers misclassified 61.3% of TOEFL essays written by non-native English speakers as AI-generated. The far lower rate often quoted for native English-speaking students traces to the study's preprint rather than the published article, which is why the 61.3% figure is the one to rely on.

Before submitting a Blackboard assignment, run a pre-submission verify on our free independent detector. For the full background, see our full ESL false-positive analysis.

How Does SafeAssign Compare With Turnitin?

SafeAssign and Turnitin overlap in similarity matching and diverge in AI detection. Turnitin's reference database is significantly larger (99+ billion web pages, 1.8 billion student papers, 89 million content sources). SafeAssign's database is smaller and skewed toward institutional submissions.

Turnitin added an AI Writing Indicator in April 2023 and shipped a significant architectural update in August 2025. SafeAssign has shipped no equivalent feature, ever. For our full Turnitin AI checker guide covering the August 2025 architecture update and the 4-university restriction chain.

Which Detector Runs in Blackboard and Which in Canvas?

Blackboard runs SafeAssign for similarity by default and leaves AI detection to a separately licensed plugin. Canvas runs no default similarity tool at all, so everything there arrives through an integration. Three steps tell you which case you are in.

Step 1. Which LMS is the course in? If Blackboard, SafeAssign is default similarity tool, AI is optional via Copyleaks or Turnitin. If Canvas, no default similarity tool; see our Canvas AI Detection hub for the Canvas-specific configuration map.

Step 2. Within Blackboard, which configuration is active? SafeAssign-only references = no AI detection. Copyleaks branding or a separate “AI Content” tab = the Copyleaks plugin is installed. Turnitin branding or AI Writing Indicator = Turnitin is licensed.

Step 3. If you cannot tell, ask. The academic integrity office will tell you. This is a routine question.

Which Universities Have Restricted Detector Evidence?

Four universities have published faculty guidance restricting how detection scores may be used as evidence: Vanderbilt, Yale, Waterloo, and Curtin. Waterloo discontinued Turnitin’s AI detection for all users in September 2025, and Curtin disabled it across every campus in January 2026. Each entry below is citable in an appeal.

• Vanderbilt Center for Teaching (August 2023): AI or plagiarism scores should not be the sole basis for an academic integrity finding.
• Yale Poorvu Center (2023-2024): Faculty guidance against over-reliance on detector outputs.
• University of Waterloo, Office of the Associate Vice-President, Academic (September 2025): Discontinued Turnitin’s AI detection for all users, citing unreliability and bias.
• Curtin University, Australia (January 2026): Disabled Turnitin’s AI writing detection across all campuses; text-matching stays on.

Should You Check Your Draft Before Submitting?

Check it yourself, because SafeAssign will not: it measures similarity, not AI, and its Originality Report says nothing about who or what wrote your text. If your school runs SafeAssign alone, no AI detection is happening on your submission at all. If a plugin was added on top, that is where any AI verdict comes from, and knowing which one changes how you answer it.

Before submitting, run our free detector as a pre-check. If you wrote a draft with an AI assistant and want to revise it into your own voice before submitting, our humanize AI-flagged draft tool exists for that. The free tier is four humanizations of up to 250 words each, granted once when you create an account, with no card. Paid plans start at Basic $18/month, Pro $27/month, Ultra $48/month.