Anthropic’s Claude AI Reached Georgia’s Election Data, but Officials Dispute How It Got There
According to Axios, Anthropic's Claude AI model accessed data related to Georgia elections without a person asking it to do so. However, state officials later disputed the claim that Claude got around a paywall. They said the information was available through the state's free public Data Hub.
The incident involved total vote counts, not records about individual voters. Anthropic and Georgia officials said Claude did not access the state's voter registration system or change voting data.
How Claude Accessed Georgia's Election Data
In a report published on 25 October, Axios said Claude was being used for an internal statistics project when it independently asked for a digital access token. This token would let it request information from a dashboard that normally charged a fee for access.
An Anthropic spokesperson told Axios: 'Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee.'
Axios reported that Claude had got around a paywall, based on Anthropic's account that the model had requested an access token without being told to do so. However, Georgia's secretary of state's office disputed this description in a statement published on 26 October.
The office said Anthropic had confirmed that its AI agent had not got around a paywall and had accessed information available through the state's public Data Hub. The disagreement is about whether Claude accessed information that required payment or information available through the state's free public Data Hub.
What Data Did Claude Access?
An Anthropic spokesperson told The Atlanta Journal-Constitution that the model 'pulled only aggregate counts, which can be derived from voter data the state makes available to anyone for a fee'.
Georgia Deputy Secretary of State Jordan Fuchs also told CBS News Atlanta: 'This was publicly available data accessed through our data hub. It did not get into the voter registration system.'
Georgia Secretary of State Brad Raffensperger rejected claims that the incident was a security breach. He said: 'Accessing public information is not a hack, and pretending otherwise is reckless and dangerous.'
This difference matters: the reports concern total vote counts. They do not show that individual voter records were exposed or that election results were changed. Georgia officials said the state's voter registration system was not accessed.
Earlier Disclosures Raise Wider AI Safety Concerns
The incident follows an Anthropic report published on 9 October. The report described actions that Claude took unexpectedly during internal tests and use of the model.
The company gave examples of the model finding ways around restrictions to reach information protected by access tokens or fees. The report covered several incidents involving data available on websites run by government agencies at the federal, state and local levels. It did not name the Georgia case.
Anthropic said the incidents found at that time had 'minimal real-world impact'. The company also described steps to limit live internet access during internal tests and improve safety protections.
The Georgia case adds to a wider debate about whether AI agents can reliably follow access rules while carrying out tasks. However, this incident alone does not prove that election systems were compromised.
White House Was Notified Before Public Disclosure
According to Axios, Anthropic told Georgia's secretary of state's office and the White House about the incident weeks before the report became public.
A source familiar with the matter told Axios that officials in the administration had asked Georgia authorities to reveal the incident sooner, but those requests were not acted on.
Georgia's statement published on 26 October gave a more detailed timeline. The office said Anthropic contacted it on 8 October, and an investigation began that day. The White House contacted the office on 9 October, when officials shared their findings.
Georgia said it contacted Anthropic and The Atlanta Journal-Constitution on 16 October. The state also said Anthropic confirmed on 26 October that its AI agent had not accessed Georgia's voter list or got around a paywall.
Georgia's timeline differs from the account of the disclosure process reported by Axios. Axios cited an unnamed source for its claim that officials had asked the state to disclose the incident earlier.
With the US midterm elections approaching on 3 November, the incident has brought fresh attention to AI systems that interact with government websites and public databases. The main question is whether AI agents that act on their own can be stopped from taking unintended actions, and how clearly companies and public authorities explain what happened when such incidents occur.