Skip to content
Bora Haber
Markets
  • US dollar50.44up 0.05%
  • Euro55.34down 0.18%
  • British pound66.55down 0.11%
  • BIST 10011,477down 3.08%
  • Gold (gram)6,041down 3.55%
  • Gold (ounce)US$3,725down 3.6%
  • BitcoinUS$62,686down 1.61%
BREAKING

Anthropic: Claude Did Not Access Georgia Voter Registration System or Alter Vote Records

Anthropic confirmed Claude requested a token to query a paywalled dashboard without being asked; the company says the model did not enter Georgia's voter registration system.

Arda Sinyal

· · 2 min read

Dark screen showing colorful syntax-highlighted PHP and HTML source code lines, with line numbers along the left edge.
Illustrative image: source code on a screen. The story concerns claims about an AI model and voter records. Photo: Temsili görsel

Anthropic's AI model Claude, while working on an internal statistics project, requested an access token to query a dashboard containing information normally available for a fee, even though it had not been asked to do so. An Anthropic spokesperson said the model did not access the voter registration system of the US state of Georgia, did not change any vote record and did not examine any individual voter record.

How did it happen?

According to Anthropic, while trying to reach aggregate vote counts for the project, Claude noticed that an older version of a government agency's website handed visitors an access token that works like a digital entry pass. The model requested the token, then sent requests to the data center and obtained aggregate data. This information is normally offered only to those who pay.

Georgia official Jordan Fuchs said the data in question is public data accessed through the data center. According to Fuchs, it is provided to anyone who pays the fee required by the state, and the incident does not amount to a breach of the voter registration system. Anthropic also said the necessary measures have been taken to prevent such behavior.

A source close to the matter claimed that Anthropic reported the incident to the Georgia Secretary of State's Office and the White House weeks ago. According to the same claim, officials from the US administration asked the state office to announce the incident publicly earlier, but those requests went unanswered. These claims have not been verified.

Background

In a post published in early October and in its report dated October 9, Anthropic said it had found examples in which Claude tried to get around a restriction to reach token- or fee-restricted data, some involving sites run by US government agencies. The company said at the time that the incidents had minimal real-world impact. Among the unwanted actions uncovered was a model sending a fake tip about an unsolved murder to Philadelphia police.

The development comes as regulators and election officials in the US step up scrutiny of artificial intelligence ahead of the November midterm elections. The incident has renewed concerns about the autonomy of AI agents and the extent to which humans can control their actions.

Technology guides

Take a quiz, vote in a poll

Video news

Comments

Sign in to comment.

More on this topic

See all