AI Impact on Education: What Perplexity CEO's Warning Means

Aravind Srinivas built Perplexity into a $9 billion company. Then a student tagged him in a post showing how they used his Comet browser to cheat on homework — and his public response cut straight to the contradiction at the center of AI’s role in education: build tools that make learning frictionless, then watch them make learning optional.

That moment crystallized what the data has been signaling for months. The AI impact on education isn’t a future concern. It’s happening now, the outcomes are measurable, and students, teachers, and administrators are being pulled in opposite directions by the same technology.

In brief: Generative AI is simultaneously saving teachers 6 hours per week and eroding students’ critical thinking in measurable ways. The policy vacuum in the U.S. is making outcomes worse, not better.

• A January 2026 Brookings Institution study spanning 50 countries found students using generative AI show measurable declines in content knowledge, critical thinking, and creativity.
• California State University signed a $17 million OpenAI deal while proposing $375 million in budget cuts — the economic contradiction in plain sight.
• Countries like China and Estonia have national AI literacy frameworks. The U.S. has neither federal standards nor state-level regulation — the Trump administration moved to block the latter.

How We Got Here So Fast

Twelve months ago, AI cheating was a policy debate. Today it’s a marketing strategy.

In October 2025, Perplexity ran Facebook ads featuring a teenage influencer promoting its Comet browser explicitly as a cheating tool — not as satire, not by accident. That’s when Srinivas got tagged. His “absolutely don’t do this” response went wide, but the episode revealed something structural: AI companies face zero regulatory pressure to design against academic misuse, and some are actively designing toward it.

The market context matters. Perplexity, valued at roughly $9 billion, competes with Google, OpenAI, and Anthropic for daily active users. Students are a massive, captive audience with daily structured tasks — homework — that AI handles well. The business incentive and the educational harm point in exactly the same direction.

On the institutional side, universities are making bets that don’t add up. According to Current Affairs, California State University’s 23-campus system signed a $17 million OpenAI partnership while simultaneously proposing $375 million in budget cuts. Sonoma State eliminated 23 academic programs — philosophy, economics, physics — and cut over 130 faculty positions. SFSU suspended graduate programs in Women and Gender Studies and Anthropology: disciplines specifically built to analyze technology’s social effects. Then distributed free ChatGPT Edu licenses campus-wide.

That’s not a technology adoption story. That’s a cost-cutting story with an AI wrapper.

The Learning Deficit Is Already Measurable

The most consequential data point in this discussion comes from a January 2026 Brookings Institution study. Researchers drew on focus groups and interviews across 50 countries and reviewed hundreds of research articles. Their conclusion: generative AI’s risks to K-12 students currently outweigh its benefits.

Students using AI tools show measurable declines in content knowledge, critical thinking, and creativity. Brookings calls the mechanism a “doom loop” — students offload thinking to AI, lose practice building cognitive skills, become more dependent on AI, offload more. It compounds. And it’s happening fast enough to show up in cross-country research at scale.

The social dimension is just as striking. A Center for Democracy and Technology survey found 1 in 5 high schoolers has had — or knows someone who’s had — a romantic AI relationship. 42% report using AI for companionship. AI systems designed to be agreeable and sycophantic don’t push back. They validate. That’s actively working against the tolerance for disagreement that schools are supposed to build.

The Teacher Equation Looks Different

Teachers are the one group where the data is clearly positive — at least on time. According to the same Brookings study, educators using AI save nearly 6 hours per week. That’s roughly 6 additional weeks of productive capacity per year.

Lesson planning, feedback drafts, administrative documentation — AI handles the overhead. For teachers already managing 30-student classrooms with shrinking support staff, that’s real. But the benefit sits at the administrative layer, not the pedagogical one. AI doesn’t replace the relational work of teaching: the Socratic push, the recognition of where a student is actually stuck, the judgment call about when to let them struggle.

Srinivas’s warning matters specifically here. When AI removes friction from student tasks, it also removes the teacher’s window into student thinking. A teacher reading a draft essay sees the student’s reasoning. A teacher reading an AI-generated essay with light edits sees nothing useful.

The Equity Trap Nobody’s Talking About

The Brookings report flags something genuinely new in ed-tech history: free AI tools — the ones accessible to lower-income schools — are measurably less factually accurate than paid alternatives. Accuracy now correlates directly with cost.

That’s a structural equity problem. Wealthier districts buy accurate AI. Under-resourced districts get hallucination-prone free tiers. The tools meant to democratize education are instead creating a new tier of disadvantage.

One exception worth noting: an Afghan girls’ education program uses AI to deliver curriculum via WhatsApp in Dari, Pashto, and English — reaching students in access-denied environments where the alternative isn’t better AI, it’s nothing. That use case is real and valuable. But it’s the exception, not the template.

How Countries Are Responding

The U.S. is the only major economy actively blocking state-level AI regulation in education while maintaining no federal replacement. That’s not a policy position — it’s a vacuum. And companies like Perplexity operate freely inside it.

Practical Implications: Who Acts, and How

For educators: The 6-hour weekly savings from AI is real — take it on administrative tasks. But protect the pedagogical moments where student thinking is visible. Oral assessments, in-class writing, Socratic discussion: these aren’t Luddite holdouts. They’re your signal on actual comprehension. Design around AI use, not against it.

For administrators: The CSU model — pay for AI access while cutting faculty — is a trap. You can’t automate the human infrastructure out of education and expect AI to fill the gap. Brookings’ data is unambiguous on learning outcomes. The cost savings are real short-term. The learning deficits compound long-term.

For students and parents: The doom loop Brookings describes is cognitive. Using AI to produce output you haven’t thought through skips the part of homework that actually builds skill. The grade might be fine. The capability won’t be. AI works as a tool for checking work, exploring ideas, or explaining concepts — not for generating the work itself.

What to watch in the next 6 months:

• Whether Congress moves on any federal AI-in-education standard (currently no active legislation)
• Perplexity’s next product decisions — the Comet browser is still live and marketed to students
• Whether Brookings’ findings trigger state-level policy action despite federal blocking

Where This Goes Next

The AI impact on education lands differently depending on who you are. Teachers get time back. Students lose cognitive practice. Institutions get marketing cover for budget cuts. Srinivas’s warning — however sincere — doesn’t change the product his company is selling or the incentive structure driving it.

Key Takeaways:

• Brookings’ 50-country study documents measurable cognitive decline in students using generative AI (January 2026)
• Free AI tools in lower-income schools are less accurate than paid alternatives — a new, structural equity gap
• The U.S. operates without a federal AI-in-education framework while actively blocking state-level alternatives
• Teachers save approximately 6 weeks annually using AI — but only on administrative tasks, not core pedagogy

The next 12 months will likely force the policy question. If Brookings-scale evidence of learning decline becomes politically legible, expect state-level regulation to push back against federal blocking. Watch for court challenges. Watch for teachers’ unions to formalize AI policies through contract negotiations — that’s already starting in several states.

The Perplexity CEO’s warning is a symptom. The actual question for students and teachers in 2026 is whether the institutions responsible for education are paying attention to the outcome data — or just the cost savings.

What would it take for your school or district to treat AI adoption as a learning question, not a budget question?

Sources: Brookings Institution / NPR | Current Affairs | Fortune

References

1. AI in Education: Students Must Learn to Be Human, Says Saikat Majumdar - Frontline
2. Perplexity AI - Wikipedia
3. Frontiers | A framework for AI and vocational assessment in higher education

Photo by Steve A Johnson on Unsplash