When Perplexity AI’s Aravind Srinivas introduced in September that college students might use the corporate’s $200 Comet browser totally free, the pitch was clear: a examine buddy that helps you “discover solutions quicker than ever earlier than.” However simply weeks later, Srinivas is having to remind college students to not let that examine buddy do all of the work.
The warning got here after a submit on X confirmed a developer utilizing Comet to finish a whole Coursera project in seconds. Within the 16-second clip, Comet breezes by means of what seems to be a 45-minute internet design project with the immediate “Full the project.” The consumer proudly tagged each Perplexity and Srinivas, writing, “Simply accomplished my Coursera course.”
The 31-year-old CEO responded to the video with simply 4 phrases: “Completely don’t do that.”
Srinivas’ terse public reprimand comes as AI seeps deeper into classrooms and tech corporations aggressively market their merchandise to college students beneath the banner of “studying assist.” Perplexity’s free-student provide joins a wave of comparable initiatives from firms like Google, Microsoft, and Anthropic—all touting their bots as tutors, examine buddies, or productiveness boosters.
However educators say these instruments are more and more getting used to bypass learning altogether. Many college students are merely utilizing AI to generate essays, ace quizzes, or automate full programs, undermining the very expertise these platforms declare to reinforce.
Comet, specifically, is nicely set as much as do college students’ work for them – it’s not your common chatbot. Constructed by Perplexity as what they name an “agentic” AI browser, it’s designed to do extra than simply spit out textual content: it might interpret your directions, take actions in your behalf, click on, fill varieties, and navigate complicated workflows. That degree of autonomy allows Comet to churn by means of assignments in seconds, but it surely additionally introduces new dangers when deployed.
Safety audits from Courageous and Guardio have flagged severe vulnerabilities. In some instances, Comet can execute hidden directions embedded in webpage content material—primarily permitting “immediate injection” assaults that override its meant conduct. One particularly alarming case, dubbed CometJacking by researchers at LayerX, lets a crafted URL hijack the browser and trigger it to exfiltrate non-public knowledge like emails and calendar entries.
In audits by Guardio, Comet was tricked into making fraudulent purchases from faux websites—finishing complete checkout flows with out human verification. It additionally mishandled phishing eventualities: when offered with malicious hyperlinks disguised as respectable requests, the AI processed them as legitimate duties.
On the similar time, Comet’s capabilities are exactly what make it so helpful in educational dishonest eventualities. Its designed to behave, not simply advise, which signifies that “learning assist” can shift into “doing the give you the results you want.” That shift is clear within the Coursera video, and it reframes most debates about AI in schooling: it’s now not nearly content material era (essays or summaries), however about automation in kind and performance.
