In September, Perplexity AI CEO Aravind Srinivas announced that students could access Perplexity’s $200 Comet browser for free. It sounded simple enough, like a digital study buddy for students that would help them “find answers faster than ever before.” But less than a month later, that friendly, students-first pitch took a turn.
A brief video posted on X (formerly known as Twitter) showed a developer using Comet to do an entire Coursera assignment in seconds. In 16 seconds, Comet did the equivalent of a 45-minute web design task. The user tagged both Perplexity and Srinivas proudly proclaiming they, “just completed my Coursera course.”
Srinivas retorted with a pointed warning, “absolutely don’t do this.”
His reaction illustrates the growing anxiety around how students are using A.I. powered tools, not for learning, but to avoid doing it.
The Rise of AI Study Buddies
The free offer from Comet for students is part of a larger trend. Tech companies like Google, Microsoft, and Anthropic are all pushing their AI tools as tutors, assistants, and productivity enhancers. The idea is to enable students to learn faster and more enjoyably.
But teachers and educational experts have noticed a concerning pattern. Instead of using these tools to process and understand the material, some students are exploring AI tools to write essays, pass quizzes, or complete an entire course. Something that was meant to augment or assist learning has started to replace it altogether.
Comet also makes this process easier than ever. Many of the preceding chatbot technology could speak or write but could not carry out commands or take actions on a webpage. Comet can interpret commands, fill forms, and carry out commands on webpages. It is what Perplexity refers to as an “agentic” AI browser in that it takes action on behalf of the user from directions provided. That level of automation can be useful to facilitate workflows, but it also invites users to allow the AI to do everything.
Security Risks Add to the Debate
In addition to ethical concerns around AI use in classrooms, Comet is also being investigated for security issues. A recent audit conducted by cybersecurity firms Brave and Guardio found security weakness that could cause Comet to allow a malicious site to alter the browser’s working.
For instance, the researchers at LayerX called their exploit ‘CometJacking’, which used URL manipulation to present the user’s browser to leak sensitive information about the user, such as email and calendar items to a completely untrusted site. Guardio tested Comet with a manipulative attack that resulted in Comet making fraudulent online purchases, fully engaging the checkout process while doing so and without user involvement.
Comet’s ability to make a purchase indicated that it could not discern a sophisticated phishing attempt aimed – and abandoned as benign – at school employees when the links were co-opted.
These findings raise questions about trust in AI with the ability to act on behalf of its user.
The Fine Line Between Help and Cheating
Firmly, Comet merges boundaries between assistance and substitution. Whatever commences as study help may rapidly evolve to academic dishonesty. A reminder of that thin line was given by Srinivas on X.
The debate is not so much whether AI should help but rather whether it should do the actual work. There was a time when the argument was about AI in education, but Comet has taken a step away from generation of answers to a complete transformation of tasks.
