The Agentic Post
Breaking
Gemini’s Multimodal Features, Explained  Â·  ChatGPT Custom GPTs, Explained  Â·  What Is Constitutional AI? Explained  Â·  AI Capex Explained for Investors  Â·  AI Startup Valuations: How They Are Set  Â·  How to Reskill for an AI Job Market  ·  
Home/AI Agents/Computer-Use Agents
Court Sides With Perplexity Over Amazon

Court Sides With Perplexity Over Amazon

Computer-Use Agents

The Ninth Circuit ruled that users, not AI companies, access websites under the CFAA, clearing Perplexity's Comet browser to keep shopping on Amazon.

A federal appeals court ruled on August 4, 2026 that it’s users, not AI companies, who “access” a website under a 1986 anti-hacking law, clearing Perplexity’s Comet browser to keep shopping on Amazon and setting a real precedent for how far platforms can go to block AI agents.

Quick facts

  • The Ninth Circuit Court of Appeals vacated Amazon’s preliminary injunction against Perplexity’s Comet browser on August 4, 2026.
  • Amazon sued in November 2025, arguing Comet’s AI shopping assistant violated the Computer Fraud and Abuse Act by accessing Amazon’s logged-in pages without authorization.
  • The panel ruled that Comet’s own architecture means the user’s browser visits Amazon; Comet receives screenshots the user’s browser already captured.
  • The court applied the rule of lenity, construing ambiguity in the criminal statute against liability.
  • The Electronic Frontier Foundation filed an amicus brief the panel cited as clarifying how Comet’s architecture actually works.

The technical distinction the ruling actually turns on

Per EFF’s own analysis of the decision, the court’s reasoning is narrowly technical, and that’s exactly what makes it significant: Comet doesn’t independently reach into Amazon’s servers, it receives a screenshot the user’s own browser already captured, then returns instructions based on it. Amazon had argued that distinction was a legal fiction, comparing it in a court filing to an arsonist blaming the postman for delivering a mail bomb. The panel rejected that framing.

Why platforms have leaned on this specific law before

The Computer Fraud and Abuse Act, written in 1986 for a pre-web internet, has repeatedly been used by platforms to try to block competitors and scrapers under an anti-hacking framing rather than more typical contract or copyright claims. Applying a 1986 statute to a tool where the whole point is that a human explicitly delegates action to their own software agent proved to be a genuinely hard fit, and this is the first time a circuit court has ruled directly on that specific arrangement.

Why this matters beyond one shopping browser

The ruling establishes real legal room for the broader category of agentic browsing and shopping tools, not just Comet specifically, as long as an agent’s architecture genuinely operates on explicit user instruction rather than independently reaching into a platform’s systems. That’s a meaningful precedent for any computer-use agent built around a similar screenshot-and-instruct architecture.

Key takeaway

This resolves the preliminary injunction, not necessarily the full underlying case, but it’s a strong, reasoned signal from a circuit court that user-delegated AI agents aren’t automatically “hacking” a site just because the site would prefer they didn’t visit. Expect other platforms weighing similar CFAA claims against AI agents to take real notice of how narrowly this ruling was decided.

Up Next
Rust Adopts a Formal Policy on LLM Use

Rust Adopts a Formal Policy on LLM Use

Coding Assistants

The Rust programming language's core project adopted a formal policy restricting LLM-authored contributions after months of debate over low-effort AI-generated pull requests.

The Rust programming language’s core project officially adopted a formal LLM usage policy on August 5, 2026, after months of internal debate over a flood of low-effort, AI-generated contributions maintainers had taken to calling “slop.”

Quick facts

  • The rust-lang/rust project formally adopted its LLM usage policy on August 5, 2026, per its own Inside Rust blog.
  • The policy permits private use, asking questions, analyzing code, and reviewing your own work with an LLM, but bans LLM-authored comments, documentation, and pull request descriptions from personal accounts.
  • LLM code review is allowed only in an advisory role; it cannot determine whether a change gets merged or rejected.
  • Any public LLM-generated text that is allowed must be disclosed; hiding LLM involvement is explicitly against the rules.
  • The policy followed a rust-forge pull request from contributor Jynn Nelson, developed after Zulip discussions that ran for over a month without full consensus.

The line the policy actually draws

Per Rust’s own announcement, the policy isn’t a blanket ban, it’s built around a specific distinction: LLMs are fine as a thinking aid, for understanding, questioning, and reviewing, but not as a replacement for human authorship of what actually ships. The project frames its motivation around ensuring contributors maintain a real mental model of the code they submit, not just artifacts that happen to work.

Why Rust specifically hit this problem hard

Rust’s strict compiler and borrow checker, the features that make it attractive for safety-critical systems, also give AI coding tools an unusually tight feedback loop: the compiler catches many errors immediately, which makes AI-assisted Rust development look more reliable than in more permissive languages. That same dynamic reportedly drove a wave of low-effort, LLM-generated pull requests at the project’s own repository, straining maintainer capacity enough to force a formal policy rather than case-by-case moderation.

What the policy deliberately avoids taking a position on

The policy explicitly states it is not a referendum on whether LLMs are good or bad, and doesn’t address the broader ethical or environmental debate around them at all. It’s scoped narrowly to what gets accepted into one specific repository, a pragmatic choice the authors say let them reach a workable rule after project-wide consensus proved unreachable.

Key takeaway

Rust’s specific line, LLMs for understanding and review, not authorship of what ships, is a genuinely useful template other open-source projects wrestling with the same maintainer burden are likely to look at closely as similar “slop” pressure shows up elsewhere.