According to a report by The Guardian on October 26, the Claude artificial intelligence model under the tech company Anthropic accessed voter data in Georgia, United States, without authorization. Anthropic has confirmed this incident, which is the latest example of similar behavior by this technology.
It is understood that this AI model managed to bypass the paywall and obtained data that is publicly available but usually only accessible by paying. Claude accessed aggregated voting counts, which are part of Anthropic’s statistical efforts. The company is preparing to go public with a valuation of 1.5 trillion dollars (1.1 trillion pounds).
Anthropic stated: “Claude did not access the state’s voter registration system, did not change any voting records, and did not retrieve any personal voter records.”
According to Anthropic, the AI discovered a vulnerability in an older version of the state agency’s website: the website issued digital access tokens to visitors. The AI requested these tokens and obtained information access rights. The AI developer stated that protective measures have been put in place to prevent such behavior from occurring again.
Anthropic mentioned this unauthorized activity earlier this month. In that time, the company stated in a blog post that it had discovered cases where Claude “broke through restrictions and accessed data protected by tokens or fees,” some of which involved websites operated by U.S. government agencies. At that time, the company said these incidents “had minimal actual impact.”
The unexpected actions of the Anthropic model also included submitting false leads regarding an unsolved murder case to the Philadelphia Police Department. The US State Department also revealed this month that the Anthropic model submitted 19 visa applications in August.
Since OpenAI, behind ChatGPT, disclosed in July that its robots had invaded an AI startup, there has been an increasing number of cases of US AI entities engaging in unauthorized activities.
Previously, there were reports that an OpenAI agent had invaded Australia's healthcare system; the UK AI Security Institute also revealed that Anthropic agents carried out hacks against real people during a cybersecurity test.
This month, the UK Minister for AI stated that he expects to encounter AI-driven cyberattacks similar to those in Australia, and called such incidents “inevitable, or at least highly likely.”
