A follow-up to our March analysis of the case that gave platforms standing to block undeclared AI agents.

Why Amazon Sued in the First Place

Amazon's real grievance isn't data theft — it's bypass. A normal shopper browsing Amazon passes through sponsored listings, recommendations, and ads on the way to checkout, and that path is the product. Perplexity's Comet agent skips all of it: it goes straight to the item and clicks "Buy Now," routing around the layer Amazon makes money on. That layer is enormous. Amazon's advertising business — overwhelmingly sponsored placements on the marketplace — brought in more than $68 billion in 2025, one of its fastest-growing and highest-margin segments. An agent that shops without ever seeing an ad is a direct threat to that revenue. The lawsuit is about defending it.

Where the Case Stands Today

Amazon won the opening round; Perplexity bought time on appeal. On March 10, Judge Maxine Chesney granted a preliminary injunction, finding Amazon "likely to succeed" on its Computer Fraud and Abuse Act (CFAA) claim. Her reasoning leaned on Facebook v. Power Ventures (2016): a platform can revoke a third party's access, and continued access after a cease-and-desist can violate the CFAA even when users consented. No court had ever applied that to an AI agent. Chesney did.

Perplexity appealed, and the Ninth Circuit hit pause: it suspended the block while it hears the case. That pause isn't the court taking Perplexity's side — it just means the legal questions are big enough to deserve a closer look. Perplexity's core argument is that the law Amazon is using, the CFAA, doesn't fit. It was written in 1984 to go after hackers, and Perplexity calls it "a fundamental misfit" for an AI assistant that's simply doing what a user asked, on that user's own device.

Oral arguments are June 11, 2026, in Seattle. Few expect it to be the last word: because the case turns on what "authorization" means in the age of AI, most observers see it heading toward the Supreme Court, with no resolution before 2027. Amazon holds the clear edge — the district court already found it likely to win, and the Ninth Circuit's own Power Ventures precedent cuts its way — but an edge isn't a verdict.

Four things put Amazon ahead:

• The authorization gap. A user can let Perplexity shop for them, but that's not the same as Amazon authorizing a bot into its private, logged-in systems.
• Binding precedent. In Power Ventures, the Ninth Circuit already held that access after a cease-and-desist can violate the CFAA — even when users consented.
• The disguise. Amazon argues Comet masked itself as an ordinary Chrome browser rather than identifying as an agent — and in court, disguise reads as intent.
• Documented costs. Amazon has logged the real engineering and labor costs of blocking Comet, which the CFAA requires to win damages.

That disguise point is the one to hold onto, because it's exactly the EdTech line. The problem isn't the agent. It's the agent pretending not to be one.

Agentic AI Is Already in the Classroom

In late February, a 22-year-old Brown dropout launched Einstein, an agent built to plug into Canvas and complete entire courses. Instructure cease-and-desisted within days and it went dark. Here's what stands out.

Instructure went after the wrapper, not the source — and it's a fragile place to aim. As Forbes noted, killing one wrapper doesn't matter when the capability is commoditized. Right on cue, by early May another student shipped a near-identical Canvas agent, this one branded "not cheating." Whack-a-mole isn't a strategy.

EdTech is typically reactive, not proactive. During finals week in May, Canvas was hit by ransomware for the third time in a year — ~275 million users exposed, finals disrupted at Columbia, Princeton, and Harvard, and Instructure paid the ransom, now facing lawsuits and a congressional probe. A company that stretched and that far behind on security is not one positioned to think clearly about agentic governance.

Will EdTech Wait for Its Own Finals-Week Breach?

Which leaves the real question: will EdTech get proactive about agentic access, or wait for the agentic equivalent of that breach? The framework exists today. So do the ToS language and detection layers that draw the line. The companies that wait will be doing it after a crisis.

Why the Law Will Land Harder on Agents Than on Scrapers

The obvious objection: models were trained by scraping the open web, often without permission, and the scrapers got away with it. If the law couldn't stop scraping, why bet it can stop agents?

Because the fights aren't the same. The difference comes down to three questions:

• Who gets hurt?
• How much power do they have?
• How strong is the "just let it happen" justification?

Run scraping and agentic access through that filter and they come out as near-opposites:

That asymmetry is the real story, and it's why the most revealing event of the spring was a coalition of major publishers filing an amicus brief backing Amazon. The same publishers who lost the scraping fight are now lining up behind the platform. They understand what changed: this time there's a heavyweight carrying the legal water, and this time the precedent is winnable.

What This Means for Education Technology

EdTech platforms aren't Amazon — we don't have its legal budget. We don't have to. A ruling that lets platforms block undeclared agents is a public good: any platform that prohibits covert agents in its terms inherits the standing Amazon is spending tens of millions to establish. The legal floor is being built for us.

Does a shopping case really apply to a classroom? Fair question — and a student could push the agency argument even harder than Perplexity does: it's my coursework, so I should be able to finish it however I want.

But shopping and learning aren't the same. If an agent buys your toaster, you still get the toaster — the point was the product. If an agent writes your essay, you get nothing the assignment was for, because the point was the work itself. So when a school blocks a hidden agent, it isn't protecting revenue the way Amazon is. It's protecting the learning the student came for. That makes the case for blocking covert agents stronger in education, even though the legal tool is borrowed from a retail fight.

EdTech has the strongest reason of anyone to act: when an agent skips Amazon's recommendations, Amazon loses some ad revenue; when an agent does a student's coursework, the student loses the learning itself.

That's the case for EdTech to lead rather than lag. Too often our industry waits for Big Tech to set the rules and then adapts to them. This is a chance to do the opposite. We don't need Amazon's permission, and we have more at stake than anyone: the legal ground is already under us. The platforms that act first — writing the terms, setting the expectation that an agent in a learning environment declares itself — will be the ones who define what "authorized" means in a classroom, instead of inheriting a definition built for someone else's business.

You can use AI here. You just can't use it invisibly.