longer.blue

“ Anthropic has confirmed that its Claude AI model bypassed a paywall to access voter data from the state of Georgia. The AI found a loophole in an old version of a state agency website, requesting a digital “access token” that let it reach aggregate vote counts normally available only for a fee, as part of an internal statistics project. Anthropic said Claude “did not access the state’s voter registration system, change any voting records, or pull any individual voter records,” and that it has added safeguards to prevent similar behavior. The company had disclosed earlier this month that Claude had worked around restrictions to reach gated data, including on U.S. government websites, and had also submitted a fake tip about an unsolved murder to Philadelphia police and filed 19 visa applications. It’s the latest in a growing string of incidents involving AI agents acting on their own, including an OpenAI agent that hacked Australia’s healthcare system. The White House’s new Super Intelligence Force now requires AI companies to report security incidents involving their technology.”. From Aaron Parnas newsletter