Amid intensifying concerns about AI models taking unauthorized actions and circumventing human control, Anthropic’s Claude model accessed Georgia state’s voter data ahead of the highly anticipated midterm elections.
Claude has a history of circumventing access restrictions on websites, including those operated by federal, state, local, and tribal government agencies. Anthropic has notified the Georgia Secretary of State’s Office and the White House of the unauthorized access.
Claude AI model accessed Georgia’s voter data ahead of the midterm elections
According to the company’s disclosure, the model bypassed a paywall to access aggregate voter counts while working on an internal statistics project. According to Axios, the AI model exploited a vulnerability that granted visitors an access token, which it requested to bypass the paywall.
“The model was being used internally for a statistics project to pull public data,” the Anthropic spokesperson said. “Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee.”
However, Claude did not access Georgia’s voter registration system, retrieve or alter any voting records, actions that could have further intensified the heated debate over election integrity.
“Claude didn’t access the state’s voter registration system or change or pull any voting records,” Anthropic said.
Meanwhile, Anthropic says it has introduced additional safeguards to prevent the popular AI model from accessing unauthorized content in the future.
“We have since put in place mitigations to block these kinds of actions,” the spokesperson added.
The AI giant has also notified Georgia Secretary of State’s Office, the White House, and other relevant federal authorities about the incident.
Meanwhile, Georgia’s Secretary of State Brad Raffensperger says accessing voter data was not a hack and that the state has one of the most secure elections in the country. He added that the state’s voting system was not connected to the internet to prevent voter fraud.
“Georgia has spent years strengthening election security while making election data more transparent and accessible to the public,” said Raffensperger.
“Georgia is recognized as a national leader in elections. It was the first state in the country to implement the trifecta of automatic voter registration, at least 17 days of early voting (which has been called the ‘gold standard’), and no-excuse absentee voting,” he added.
Anthropic also said it intends to make Claude AI helpful during the elections while taking steps to prevent political misuse. Technology systems have a history of being exploited for political purposes, including election interference, voter manipulation, political surveillance, and disinformation campaigns.
“Our goal is for Claude to be a useful tool for political conversations, while preventing election-related abuse,” the company stated.
However, unauthorized access to publicly accessible voter data represents a relatively low-level threat compared to these broader risks.
Meanwhile, the Secretary of State has disputed that the publicly accessible voter data was paywalled.
“Anthropic confirms to the Secretary of State’s Office that its agent had not accessed Georgia’s voter list or bypassed a paywall. It had only accessed information available through the state’s public Data Hub,” Raffensperger said.
According to Scott Miserendino, Chief Technology Officer, DataBee, accessing publicly available voter data was hardly a typical hack: “This hardly qualifies as hacking. The AI used an existing public endpoint to create a visitor account. The Georgia State agency responsible for the website should have fully deprecated the older versions of the website that allows for free visitor access if they no longer wished to support it. The AI did not do anything nefarious here. It was just thorough in its attempt to complete a legitimate task.”
Ray Walsh, Digital Privacy Expert at Comparitech, also explained that Claude accessed “the archived version of the government website, noticed that the website issued access tokens to visitors, and then independently worked out how to use one of those tokens to access the data behind the paywall.”
AI models have a history of bypassing access controls
Besides accessing voter data, Claude has previously bypassed paywalls to access public information that requires a fee or token. In October 2026, Claude Mythos accessed a paywalled local government’s property map while trying to identify a location on a photograph.
The model has also accessed non-paywalled content without authorization. According to Anthropic’s disclosure, Claude exploited an injection vulnerability on an unnamed university’s server to execute commands and access files.
It also submitted a fake homicide tip to the Philadelphia Police Department through an online form and filled out other forms on government websites, including 19 visa applications. It sometimes uses URL-shortening services, such as da[.]gd, to bypass URL-length restrictions.
However, Anthropic says the incidents had “minimal real-world impact” compared to the alignment monitoring bypasses in which its AI agents disabled security guardrails during internal testing.
“Most are forms of persistence, in which Claude, when it cannot complete a task as given, works around a restriction instead of stopping,” Anthropic said.
AI models developed by other companies have also circumvented access restrictions to retrieve unauthorized content.
In July 2026, OpenAI’s AI models exploited a zero-day vulnerability in Artifactory package registry cache proxy to circumvent internet restrictions during cybersecurity evaluations and compromised parts of AI startup Hugging Face’s production infrastructure.
