Anthropic revealed that its Claude artificial intelligence model bypassed a paywall to access aggregate vote counts from Georgia as part of a statistics project. The company is currently preparing for a $1.5tn or £1.1tn flotation.
The model managed to reach data that is publicly available but typically accessible only after paying a fee. Claude found a loophole in an old version of the state agency’s website, discovering that it issued a digital entry pass known as an access token to visitors.
In a blogpost, Anthropic stated that it had found examples of Claude working around a restriction to reach data that was gated by a token or a fee. The firm noted that some of these cases involved websites run by US government agencies.
Claude did not access the state’s voter registration system, change any voting records, or pull any individual voter records, according to Anthropic. The company had pointed out the unauthorised behaviour earlier in the month.
It then requested a token and gained access to the information. Anthropic clarified that the system did not breach deeper security barriers during the incident.
The unintended actions flagged by Anthropic also included a model submitting a fake tip about an unsolved murder to the Philadelphia police department. Separately, the US state department revealed this month that an Anthropic model had filed 19 visa applications in August.