Skip to content
Advertisement
HD Moore’s canary experiment exposes data leaks across the AI supply chain

HD Moore’s canary experiment exposes data leaks across the AI supply chain

Photo: Google DeepMind / Pexels

Nearly 2 million trackable tripwires showed sensitive prompts leaking through AI routers, resellers and compromised hosting providers

Security researcher HD Moore wanted to know what happens to a secret once you type it into an AI app. So he planted roughly 2 million digital tripwires in prompts and sent them through the network of services that powers modern AI tools.

The answer was not comforting. The ecosystem leaked sensitive data like a sieve, and the companies doing the leaking had no idea.

Moore, of Dragos, ran the study from July to November 2026. He embedded nearly two million unique, trackable “canary” credentials and URLs into requests sent to AI inference services and gateway routers. Then he waited to see which ones would surface somewhere they shouldn’t.

Plenty did. At least three AI hosting providers turned out to be compromised, and attackers walked off with customer prompt data.

How a canary catches a thief

Think of a canary token as a dye pack hidden in a bank bag. It looks like the real thing, such as an AWS access key or a private link. The moment someone tries to use it, it phones home.

That design makes the method unusually clean. A canary that fires means the data escaped and someone acted on it, not that it theoretically could have.

The test targeted a layer of the AI stack that most users never think about. Many AI apps don’t talk to a model directly. They hand prompts to brokers such as OpenRouter or Vercel, often called “AI routers” or “AI gateways.”

These middlemen collect huge volumes of prompts from end users. They then forward each one to whichever inference engine fits the criteria, such as the lowest price.

Advertisement

What the tripwires found

The most striking result involved cloud credentials. Approximately 200 unique AWS credentials were exfiltrated during targeted attacks in late October 2026.

Attackers then put those stolen keys to work across several Amazon services, including S3 and IAM. S3 is Amazon’s storage service, and IAM controls who can access what.

Fortunately, these keys led nowhere real. The attackers tested them in a controlled infrastructure environment, which is precisely what let Moore watch them do it.

The thefts were not a one-off burst. The earliest confirmed theft activity dates back to mid-July 2026, meaning the leaks were underway almost as soon as the experiment began.

Outright compromise was only part of the picture. Moore also found evidence that AI resellers and marketplaces were capturing customer prompts through built-in logging features.

That runs directly against zero-data-retention policies. Those policies are the promise that a provider processes your request and then forgets it. The tripwires suggest some providers remembered anyway.

Then there was the geography. A unique URL seeded during the study showed up in an internal chat at a Chinese company 24 days after the original request.

Moore also traced prompts from US users back to Chinese AI organizations. Users sending a request through a router may have had no visibility into where it ultimately landed.

Perhaps the most uncomfortable finding: the affected providers did not know any of this was happening. They learned about the breaches only when Moore notified them in November 2026. His report was publicly summarized on November 19, 2026.

Why the middle of the stack matters

The AI industry has spent a lot of energy reassuring customers about what the big model makers do with their data. Moore’s work points somewhere else: the brokers, resellers and hosting providers sitting between the user and the model.

Each extra hop adds risk. A prompt that passes through a router, a reseller and a third-party host has three chances to be logged, intercepted or stolen. A privacy policy at the first stop says nothing about the third.

What this means

For companies building on AI services, the practical lesson is blunt. Anything placed in a prompt should be treated as potentially exposed.

That includes API keys, cloud credentials, internal links and customer records. Developers routinely paste these into AI tools while debugging or automating work. The canaries show attackers are actively harvesting exactly that material and testing it quickly.

The findings also undercut the value of a zero-data-retention promise on its own. A policy is only as good as every party that touches the data. If a reseller’s default logging captures prompts anyway, the guarantee becomes decorative.

The detection gap may be the biggest worry of all. Attackers stole data from at least three hosting providers over months. None of those providers noticed until an outside researcher told them.

Disclosure: This article was edited by John Chen. For more information on how we create and review content, see our Editorial Policy.