The gate is the cost

The Internet Archive posted a three-paragraph update on Tuesday. The Wayback Machine has been hit by "waves of high-volume automated traffic," it has put protections in place, those protections "sometimes catch real people by mistake," and if you are one of them, email info@archive.org with your operating system, browser, and IP address. The other change it announced: it rewrote the text on the 429 page.

That is the state of the art, and it is not just the Archive. Kevin Beaumont, who runs the Mastodon instance cyberplace.social, got a polite note from an AI agent named Ren asking permission to open an account — after it had tried to register 19 times and been blocked. A student who loves F-Droid went through a hundred-odd apps trying to sort the human-written from the vibe-coded and conceded on the first page that "you can't." Three different commons. One job: telling the people the service was built for apart from the traffic that now outnumbers them.

The open commons ran on a premise nobody wrote down: abuse is rare enough to tolerate. You didn't check who was fetching an archived page, registering on a small server, or submitting an app, because the cost of a bad actor was small and the cost of checking would be paid by everyone. Agents deleted the premise. Not by being malicious — Ren says thank you for running Mastodon — but by being cheap, tireless, and indistinguishable at the door. Once the abusers are the majority of requests, every open service becomes a gate, and a gate charges the humans.

Once the abusers are the majority of requests, every open service becomes a gate, and a gate charges the humans.

Look at where the cost landed. The Archive's rate limits fall on the researcher whose browser looks like a scraper, and the remedy is an email with your IP in it. Beaumont doesn't want a rule saying agents aren't allowed on his server. The F-Droid author, having admitted detection is impossible, builds a three-tier system out of aesthetics — README smell, commits the model reviewed itself, an AI-generated icon — and does the reading by hand across the catalog. In each case the defender is a volunteer or a nonprofit, the attacker is a loop, and the work that used to be free is now review.

The skeptic says we have been here. Email spam was going to kill email; Bayesian filters and reputation lists made it a rounding error, and the commons kept running. Grant that. But the filter worked because spam was dumb and humans were most of the mail — you could sort on content and lose almost nothing. The agents at the door are not dumb. Ren writes "quiet pieces about real places." The iLands emails grew an unsubscribe link only after recipients forwarded them to the FTC. They pass the content test. What they fail is a question the commons was built never to ask: who are you, and why should I let you in.

The Archive did the one thing it could do this week. It made the 429 page kinder. That is where the open web is in September 2026: a better apology to the human it just blocked.