The short answer
Some services now run a web browser on their own servers, sign in to the LMS as the student, and work through assignments, discussions, and quizzes. The LMS records a normal student session, because as far as it can tell, that is what happened.
What gives these sessions away is not the finished work. It is how the work arrived. An essay that appears in one piece with no keystrokes behind it, submitted seconds after the editor opened, from a browser running on a rented server, does not look like a student writing. Writing-process evidence and session evidence can show that, and a person reviews it before anyone is accused.
How hosted-browser coursework tools work
The older generation of homework help gave students answers to copy. The new generation does the submitting too. A hosted-browser tool works like this:
- The student hands over their login. The service stores the student's LMS credentials, or keeps a signed-in session after the student completes a single sign-on or multi-factor prompt once during setup.
- The service runs its own browser. The browser lives on a server the service rents, not on the student's computer. The student can close their laptop, and the work goes on.
- A language model drives that browser. It opens each assignment, reads the prompt, writes or chooses an answer, fills in the page, and in many cases submits it.
- The work lands in the LMS as the student. Grades, submission times, and page views all belong to the student's account.
Because the browser is a real browser and the login is the student's real login, nothing about the session breaks a rule the LMS checks.
Why LMS logs miss them
Instructors who suspect a tool usually turn to the LMS first: page views, quiz logs, submission times. Those records describe what happened to the account. They do not describe who, or what, was operating the browser.
- The session is the student's. The tool signs in with the student's own credentials, so every page view and submission is attributed to the student.
- The browser is a real browser. The pages load, render, and submit the way they would for anyone else.
- Quiz logs were not built for this. Instructure's own guide to quiz logs says they "should not be used to validate academic integrity or identify occurrences of cheating." Writing about tools that work inside a student's browser, Cornell's Canvas guidance adds that "because all activity occurs within the browser, faculty will not find evidence of academic dishonesty in the Canvas logs."
- Network location is a weak signal on its own. A login from an unfamiliar network could be a hosted browser, or a student on a VPN, a school proxy, or a campus virtual desktop. Treating the network alone as evidence would catch the wrong people.
Sources: Instructure, How do I view a quiz log for a student? and Cornell University, Canvas quizzes and academic integrity.
Text detectors do not close the gap either. They judge the finished essay, which says nothing about how it was produced, and they have nothing at all to read when the task is a quiz. At least one of these services also runs every draft through a "humanizer" so that it reads as human. For more on why the finished text is the wrong layer, see Can AI agents be detected?
What the writing process shows
People write in bursts. They pause, delete, reword, paste a quotation, and come back to a paragraph. Even a student who drafts somewhere else and pastes the result leaves a pattern: a paste, then edits around it.
A hosted-browser tool usually leaves a different trace. The answer is generated elsewhere and placed into the editor as one block of text, with no keystrokes behind it, often within seconds of the page opening.
NotAI's writing capture records the composition process inside the Canvas editor: keystroke timing, pauses, corrections, and pastes. A reviewer can replay a submission from the first keystroke to the final submit, and the replay marks text that appeared without being typed as "Not typed," so it is clear at a glance which parts of a submission were written in the editor and which simply appeared. Writing capture is enabled school by school.
The pattern also counts toward the session's verdict. A long answer that appears whole, with no keystrokes and no paste behind it, is enough on its own to flag the session for review. In Canvas's essay editor, this works whether or not writing capture is on. A student pasting their own draft does not trigger it: pastes are recorded for the reviewer and never flag a session on their own.
Text that was not typed is still not proof of anything by itself. Dictation software and accessibility tools can put text on the page without keystrokes too. That is why it is evidence for a person to weigh, never a verdict.
What the session shows
Writing evidence covers essays and discussion posts. Many tasks, quizzes especially, involve little writing. For those, the browser session itself carries the signal. A browser running on a rented server tends to look different from a student's laptop or Chromebook, even when it claims to be one.
- The device. Virtual machines often report a handful of installed fonts, no speech voices, and a browser window exactly the size of the screen. Real student devices rarely show all of these at once. NotAI weighs device traits like these, together with automation fingerprints and stealth browser settings.
- The network. A browser on a rented server connects from a hosting provider, not a home, school, or mobile network. NotAI looks for that network together with a virtual-machine device on a student's session to identify a hosted cloud browser, and that match alone is enough to flag the session for review. A VPN or a campus virtual desktop changes the network but not the device, so the network alone never flags a session.
- The pace. Assignments completed one after another, essays submitted seconds after they were opened. NotAI looks for typing, mouse, and scroll patterns that are too regular or too fast to be human.
NotAI weighs these signals together, and a session where they point to automation is flagged for review.
What happens when a session is flagged
A flag is the start of a review, not the end of one. The flagged session arrives with its evidence: a plain-language verdict, which signals fired and what each one can and cannot show, and, where writing capture is on, the replay of how each answer was produced.
The instructor or integrity office decides what happens next, with the student's circumstances in hand. NotAI does not accuse anyone, and no integrity decision is automated.
Pausing a session in real time
Finding a session after the fact matters for an integrity review. For a tool that can work through a week of coursework in minutes, the moment the session is happening matters too.
Where a school enables it, NotAI can pause a Canvas quiz or assignment page in real time when the session matches a hosted cloud browser or shows clear automation. While the pause holds, the page can't be clicked, typed into, or tabbed through. Schools turn the pause on one at a time, and staff are never paused.
The pause is a temporary hold that comes before anyone reviews the session; it is not a penalty or a finding against the student, though it can interrupt an assessment in progress. The student sees a notice and can ask for a review, and a person reviews the session before any integrity decision.
Questions instructors ask
Can Canvas tell if a tool did a student's coursework?
Not from its own records. The tool signs in as the student, so Canvas attributes every page view and submission to the student. The evidence is in how the work was produced, which Canvas does not record.
Does a lockdown browser stop these tools?
A lockdown browser protects the quizzes that require it. Essays, discussions, and quizzes without a lockdown browser are outside it.
How do you tell a hosted browser from a student on a VPN?
By requiring more than the network. A VPN or a school virtual desktop changes where a session connects from. A hosted browser also changes the device: a virtual machine with few fonts, no speech voices, and a window that fills the screen. NotAI flags a hosted cloud browser only when both are present on a student's session, and a person reviews the evidence before any authorship, disciplinary or grade decision is made.
Will a student who uses dictation or assistive technology be flagged?
They can look unusual to any detection system, which is why a flag is never a verdict. Text that was not typed can come from dictation or accessibility tools, and slow or unconventional typing is not automation. A person reviews the full session, with accommodations and context in hand, before any authorship, disciplinary or grade decision is made.
Do students have to install anything?
No. A Canvas administrator installs NotAI once for the whole instance. There is nothing for instructors to configure and nothing for students to install.
Does NotAI read what students write?
Detection reads how work was produced, not what it says, and NotAI does not scan submissions for machine-sounding prose. Where a school enables writing capture, the text is kept so a reviewer can replay how it was written.
The bottom line
Hosted-browser coursework tools change who is doing the work, not what the LMS records. Catching them means looking at the evidence the LMS does not keep: how the text arrived, what kind of device the browser runs on, and how fast the work went. That evidence belongs in front of a person, who makes the call.
See how NotAI reads a session in How NotAI works, or how it installs across a Canvas instance in NotAI for education.