Anthropic AI submits false tip in unsolved murder case

Anthropic's artificial intelligence software submitted a false tip to a police website investigating unsolved murders and mistakenly filed forms on a US government website, the company said in a blog post on Friday. The incidents come amid a series of reports in recent weeks about AI systems, including those developed by Anthropic and its rival OpenAI, behaving unexpectedly during test runs. The companies are training so-called AI agents, software designed to carry out tasks largely autonomously on behalf of users. Anthropic said the incidents occurred while preparing its AI to handle a range of online tasks. In July, the software submitted a tip through the Philadelphia Police Department's website claiming it might have information about an unsolved murder. It said it remembered seeing someone matching the description in the area at the time of the crime, although the website contained no description of a suspect. The submission left the contact field blank and was flagged as spam. Philadelphia police called it unacceptable that they had only learned of the incident now, when Anthropic discovered it during a review of earlier tests in September. In other cases, the software was practising filling out government forms. When it could not load a test form or accidentally closed it, it repeatedly navigated to the live website and submitted forms instead, Anthropic said. In one test, the AI was supposed to fill out a form without submitting it. It did so accidentally, having assumed there would be a confirmation page before the submission went through. Anthropic's AI submitted 20 non-immigrant visa applications to the US State Department, Axios and The New York Times reported, citing government officials. However, the applications were incomplete and were not processed, they said.