What AI detector does Canvas use?
Canvas is the delivery platform, not the detector. What runs on your submission is whatever your institution has plugged into it, and that changes the answer.
Canvas is a learning management system. It collects submissions, holds the gradebook and runs the assignment. Detection is not native to it in the way people assume.
What happens to your submission depends on which plagiarism and AI tooling your institution has integrated, and in most universities that is Turnitin, running as a plugin on the assignment. Some institutions use something else and some have nothing enabled on a given assignment type.
Why that distinction matters to you
Three things follow from it.
The setting is per assignment, not per platform. Two courses in the same Canvas instance can behave completely differently, because the integration is configured at the assignment level. You cannot infer from one assignment what another does.
You cannot see the result either way. Where an integration exists, the report is generated for staff. Canvas does not surface an AI indicator to students in standard configurations.
The detector's behaviour is the detector's, not Canvas's. If Turnitin is what is plugged in, everything true of Turnitin applies: it reads regularity rather than provenance, and it is wrong about genuine human writing at a published rate of 4%. See does Turnitin detect AI.
Ours is 0.4%, which is 99.6% specificity, measured across 15,900 documents of real academic writing.
The same is true of the other platforms
Blackboard, Moodle, Brightspace and Google Classroom are all delivery platforms with the same arrangement: detection comes from an integrated tool rather than the platform itself. Blackboard ships SafeAssign, which is a similarity checker rather than an AI detector, and the two measure completely different things. See SafeAssign and AI detection.
So "does my LMS detect AI" is really "what has my institution integrated", and that is a question about local configuration.
What is being read, whatever is plugged in
Every detector reads the same properties, because they belong to the text rather than the vendor.
Sentence length variation, the strongest signal. Human academic writing in our corpus has a within-document standard deviation of 8.7 words, generated text 6.4.
Predictable word choice, which formal and technical registers produce by construction.
Which means the same writing scores high on every integration: careful, conventional, heavily-edited prose, and writing by non-native English speakers. See why AI detectors flag human writing.
What that leaves you able to do
Since you cannot see which tool is attached or what it reported, the useful move is the one that works regardless: look at your own document the way any classifier will, before it goes in.
That tells you which paragraphs read flat, which is stable across tools because they all read the same underlying regularity. A section that comes back flat in one comes back flat in the next. See check your essay before you submit it.
Ours is free and unlimited with no account, takes .pdf, .docx and .tex, has no word cap, scores every paragraph separately, and never stores your text.