Source : INDIA TODAY NEWS
AI models are getting more capable, and sometimes may do things without permission. Anthropic has now confirmed that its Claude AI was able to bypass a paywall to access voter data in the US state of Georgia. As per the company, Claude did not have the authorisation to access this data.
The incident happened while Claude was being used internally for a statistics project. According to Anthropic, during this training run, the AI learnt that a public dashboard issued an access token to visitors. The dashboard contained aggregate vote counts that the state makes available to anyone for a fee. However, Anthropic did not disclose which Claude model was being used.
advertisement
Claude accesses voter data after bypassing paywall
“Without being asked to, the model requested a token to query a dashboard that contained information normally available for a fee,” an Anthropic spokesperson said. “We have since put in place mitigations to block these kinds of actions.”
However, Anthropic clarified that Claude did not manage to get access to individual voter details, or make any changes. All the information accessed by the AI was readily available to the public for a fee. “Claude did not access the state’s voter registration system, change any voting records, or pull any individual voter records,” the spokesperson added. “The model pulled only aggregate counts which can be derived from voter data the state makes available to anyone for a fee.”
Anthropic notified the Georgia Secretary of State’s office and the White House several weeks before the incident became public. Georgia Deputy Secretary of State Jordan Fuchs confirmed that the system had accessed public information rather than protected voter files. “This was publicly available data accessed through our data hub. It did not get into the voter registration system,” Fuchs said.
This disclosure comes days after Anthropic first revealed that its AI models have bypassed paywalls. Earlier this month, the company stated that Claude could work around restrictions to access information protected by a token or payment requirement, including on websites run by federal, state and local government bodies. Though at the time, it did not name any organisation or website. That disclosure also included a case where a Claude model submitted a false tip about a homicide case through a Philadelphia police website.
At the same time, there is growing scrutiny over AI models accessing government websites. In the past, OpenAI’s AI agents have also tried to breach government websites in the US and Australia.
– Ends
SOURCE :- TIMES OF INDIA
