Claude's crawler in Georgia: public data, but the real debate is what counts as authorized access

By FOX 5 Atlanta (with reporting from Reuters and Axios) · October 25, 2026.

An artificial intelligence bot associated with Anthropic accessed Georgia voter data through the state's online data hub, election officials confirmed. Deputy Secretary of State Jordan Fuchs said: “This was public data accessed through our data hub. It did not enter the voter registration system.”

According to FOX 5, the automated program interacted only with public data sources. The state says protected voter registration files and non-public records were not affected.

Outlets differ in how they frame the story. Axios, according to the headline picked up by aggregators, says Claude “got around a paywall” to access the data without authorization. Reuters, citing Axios, describes it as access to public voter information. The available material does not let us verify the paywall claim or how the access happened. We attribute it to Axios and treat it as an unconfirmed allegation.

The key facts are also still unknown. Officials have not released the date or duration of the crawl. It is also unconfirmed which research or training project the downloaded files were used for. So far, the material we have includes no statement from Anthropic.

**Our take.** The apparent harm is small, but the precedent is significant. Data being public does not answer the question that matters for an AI agent: under what conditions can it access and reuse that data? If a paywall, terms of use or an access control is involved, the content being “public” is no longer a sufficient defense.

More broadly, the web is splintering into walled gardens. Platforms such as Amazon, Delta, JPMorgan and Schwab already block AI agents, and what an agent can reach is increasingly decided by site owners, not by the model. This episode adds a new twist: government agencies will also have to decide how they treat automated crawlers.

There is a political dimension too. This is election data, in the middle of the U.S. election season. Even though the state rules out any breach, the context means any automated access to that data immediately raises suspicion. That is one more reason for AI companies to be transparent about what they crawl, under what identity and for what purpose.

In the short term, this means more friction: agent identification, access agreements and probably litigation over what counts as “authorized.” In the long term, we think that friction is healthy. Clear identity and permission standards for agents lay the groundwork for them to eventually operate at scale on public services and improve them, without every crawl turning into a scandal.

The test will be simple: what Anthropic says, what details Georgia provides about the access, and whether the alleged paywall existed. Until then, it is worth separating what is confirmed (access to public data, no intrusion into the registration system) from what is attributed to a single outlet.