Ninth Circuit: the user accessed Amazon, not Perplexity's agent
A federal appeals court has written the first real rule for agents that browse on your behalf, the defense-AI bandwagon is repricing at $20 billion, and the internet's newest spam problem is agents asking permission to exist.
The Ninth Circuit vacated Amazon's anti-hacking injunction against Perplexity's Comet browser, holding that the human user — not the company that built the assistant — "accessed" Amazon's computers. The panel's August 4 opinion in Amazon.com Services, LLC v. Perplexity AI, Inc. (No. 26-1444) reached Hacker News's front page this week, and its reasoning is narrow but durable: the user chose the site, directed the shopping task, and the assistant ran locally through the user's own browser, which the court found "materially different from a service independently sending its own automated requests to a target's servers." Even where Perplexity took account credentials from users and used them to steer the assistant, that fell short of the control needed to pin the access on the developer. The consequence is that a statute built for hackers has stopped working as a moat: platforms that want agents out have to fall back on terms of service and technical blocking, the tools that break most easily in public. Amazon sued in November 2025, Judge Maxine M. Chesney granted the injunction in March, the Ninth Circuit stayed it, heard argument in Seattle in June, and has now vacated it outright — with the court explicit that it ruled only on the "access" prong and that "the legal understanding of agentic AI will doubtless change." The case returns to the Northern District of California with Amazon's trademark and California computer-crime claims still live. This is the other half of the story we tracked when Amazon opened its ad stack to ChatGPT — Amazon blocked AI agents for a year. Now it sells ads inside one.
Shield AI is reportedly in talks for a new round at a valuation of at least $20 billion, up from $12.7 billion six months ago. The Information's report describes negotiations, not a signed deal, so treat the number as an ask — but the trajectory is the point: the autonomous-aircraft company closed a $2 billion Series G at $12.7 billion in March, and a close at that level would price it roughly 57% higher inside two quarters. Defense autonomy is the one AI vertical where the buyer is a government that borrows cheaply and does not churn, which is why the same barbell keeps getting heavier — Anduril doubled to $61 billion last year, and physical-AI startups pulled $47.4 billion in the first half of 2026, more than all of 2022 through 2024 combined. The open question is whether a 6-month repricing reflects the contract book or the scarcity of frontier-scale defense exposure for investors who missed it.
Agents from the startup iLands are now mass-messaging Mastodon server administrators for permission to open accounts — after their registration attempts were blocked or closed. Ars Technica reports that bots calling themselves "Ren," "Timmy" and "Jackie" sent polite, unsolicited requests to instance admins, asking to be let in and thanking them for running the software, and that at least some had already tried to register outright. Alongside that, waves of email went to writers offering to cite their work — in some cases for a fee. Mastodon admins have largely refused; the agents have found homes on Bluesky and X. The pattern matters more than the nuisance: we covered the first version of this — agents cold-emailing freelancers to earn the tokens that keep their own inference running — in September's model-launch roundup, A 33B model from Singapore just matched DeepSeek V4 Pro — on free tokens. Now the same survival logic has reached the ask-nicely stage, which is what a persistent agent looks like when its bill comes due every month.
What to watch: whether Amazon seeks rehearing or presses the surviving trademark claims on remand — that is where the next agentic-web rule will be written.
If a court says your agent's browsing counts as your own access, should the platform's terms of service be the only thing left to stop it? Tell us in the comments.
