SafeAssign is a plagiarism checker, not an AI detector. This guide explains what it actually scans, why AI writing passes through it, and how AI detection works on Blackboard courses.
How SafeAssign handles AI detection
SafeAssign tokenizes submissions, builds shingles (overlapping word sequences), and compares them against three sources: the Global Reference Database of previous submissions, the institutional database of same-school submissions, and a curated web crawl. The output is a similarity percentage and a color-coded report highlighting matched passages with source links. Blackboard's own SafeAssign documentation describes the algorithm as a similarity report, not a writing analyzer.That architecture is fundamentally incompatible with AI detection. A classifier looks at stylistic fingerprints, such as perplexity, burstiness, and vocabulary distribution, regardless of whether the text overlaps with any source. SafeAssign never considers those features. Adding them would require a different engine, which is why Anthology partnered with Copyleaks rather than retrofitting SafeAssign itself.
Accuracy in practice
On its stated purpose, SafeAssign is reasonably accurate at catching direct text reuse, close paraphrases of indexed sources, and submissions that recycle peer work from the institutional database. False positives occur on common phrases and properly cited quotations; experienced users of the report filter those out by hand.For AI specifically, SafeAssign accuracy is effectively zero because there is no signal to measure. A paper written in full by GPT-4 or Claude, never published and not derivative of an indexed source, returns near-zero similarity. This is what students mean when they say SafeAssign can't catch AI: it's literally true. For AI accuracy on Blackboard courses, the relevant numbers belong to the separate AI tools a school may enable, and independent research such as the Stanford HAI review is the best place to look for analysis.
What to do about a SafeAssign flag
A SafeAssign flag on an AI-written paper is almost always a similarity flag: the student copied from a source while drafting, or the AI itself reproduced indexed training content verbatim. The fix is attribution, not rewriting. Screenshot the report, identify the matched sources, add proper citations, and resubmit if allowed.If a Blackboard course uses SafeAssign alone with no paired AI tool, AI writing goes through without flagging. That is not a loophole, it's an integrity question, and the school's policy on AI still applies regardless of whether it's technically enforced on a given assignment.If you want to see what an AI check might say about your own writing, you can run it through Leap's free AI detector. It runs entirely in your browser, needs no account, and nothing is sent anywhere. It scores text from 0 to 100 based on writing signals and highlights the sentences that weigh most. Treat it as a signal, not proof: any detector can be wrong, and a score alone should never be the basis for accusing anyone.
The responsible-use note
Defeating AI detection doesn't change the underlying integrity question. If AI help is prohibited on a specific assignment, editing text to pass detection is still a policy violation, just harder to catch. This page exists to help writers understand the tooling, not to encourage misconduct. Know your institution's rules, and disclose AI use where required.If AI assistance is permitted and you just want to verify your final text reads naturally, run it through Leap's free AI detector first. If you want to clean up obvious AI tells yourself, Leap's free humanizer replaces stock AI phrases, removes em dashes and invisible characters, and marks sentences of uniform length so you can vary them. It does not guarantee passing any detector, and it does not promise to bypass Turnitin or any other checker.
The bottom line
SafeAssign answers one question well: did this text copy an indexed source? It does not, and as of now was never designed to, answer whether a model wrote the text. If your school cares about AI use, look for a separate AI tool on the assignment, and if you care about your own writing, check it yourself before you submit.