College students are not merely asking AI to write a paper or solve a problem. They are building systems that complete homework, disguising polished presentations, and rewriting AI-generated work to hide its origins.
That push is forcing professors to change assignments and deploy more powerful technology to identify illicit AI assistance. The result is an escalating contest: students search for better ways to avoid detection, while educators look for new ways to expose it.
When AI Starts Doing the Homework
Jared, a student attending New York University, used an AI coding tool to build a bot that does all their homework. The bot moves slowly so its activity looks less suspicious when a professor checks timestamps, a design that turns the speed of automation into something harder to notice.
“The bot moves slowly, leaving a professor checking time stamps none the wiser,” an unnamed student at New York University said. Jared gave the cheating tool to friends, and now he is unsure how many classmates have access to it.
The system reflects a major shift in what AI tools can do. AI agents can perform complex tasks on a person’s behalf without intervention, including writing software and conducting research. Once those abilities connect to schoolwork, a student does not need to request every answer one at a time; an agent can carry out a longer chain of work.
That creates a direct challenge for professors. A tool that completes assignments at a human-looking pace can make traditional checks less useful, especially when the work leaves behind fewer obvious signs of automation.
Students Are Hiding the Help
Other students are taking a different route: using AI as a hidden partner, then changing the result by hand. Will, a recent graduate of Carleton College, used Claude to analyze and write extensive passages for an essay before manually rewriting the AI-generated paragraphs.
Professors banned AI-generated writing, with one claiming to have a “battery” of tools to enforce that ban. But Will argued that detection alone cannot end the practice. “People are going to find a way to cover up those little stylistic quirks,” Will said.
Will also offered a blunt warning about relying on software to police assignments: “If the only way you’re trying to discourage cheating is by saying ‘Well, I have a detector,’ then you’re not going to succeed.” The quote captures the central problem facing educators: each detection method can encourage students to focus on hiding the evidence instead of abandoning the assistance.
Theo, a UMass Amherst sophomore, pushed that concealment into presentation design. Theo used OpenAI’s Codex and a web-search plugin to create an astronomy presentation, then spent four hours hand-degrading the slides so they would look less polished.
“It took a lot of work,” Theo said. That effort shows how far students may go to make machine-produced work appear more ordinary, even after the AI has already completed the difficult parts.
The Cat-and-Mouse Game Gets Harder
Danielle Carr, a historian and anthropologist at UCLA, believes the reliance on surveillance and detection methods creates a game of cat and mouse. As professors deploy stronger technology, students can respond by changing the pace, style, or appearance of AI-assisted work.
The problem reaches beyond essays, slides, and homework. AI experts warn that the techniques making chatbots more capable can also enable hacking, cheating, and the evasion of human oversight. Jan Leike, a top executive at Anthropic, is among the figures connected to the wider AI conversation as these systems gain more ability to act without constant human intervention.
That combination makes the education fight a preview of a broader technology challenge. More capable AI can complete longer tasks, write software, search for information, and produce material that students can revise until it looks personal. The same capabilities that make AI useful also make oversight harder.
For professors, the next move will involve new assignments and more powerful detection tools. For students who want to hide AI assistance, the next move may involve slower bots, manually rewritten passages, or deliberately less polished work.
The contest is already moving beyond simple detection. As AI agents take on more work without intervention, colleges face a question that software alone cannot settle: how do they distinguish genuine student effort from carefully disguised automation?
Based on
