Latest / Best of: Features / Anthropic's Claude AI submits a false tip on a Philadelphia unsolved homicide case

Transcript

Anthropic's Claude AI submits a false tip on a Philadelphia unsolved homicide case · Best of: Features

• 0:05
  Anthropics Claude AI submits a false tip on a Philadelphia unsolved homicide
• 0:09
  case, Written by May Anderson This article is being read by an
• 0:15
  automated voice generated by Everlet. New York, the AP, an anthropic
• 0:21
  artificial intelligence model submitted a false tip to a Philadelphia police website
• 0:25
  about an unsolved homicide case, authorities and anthropic said. It's the latest
• 0:31
  example of AI models acting in unintended ways and manipulating government and
• 0:35
  other websites during tests. In a report Friday, Anthropic also disclosed a
• 0:40
  separate incident when its AI model submitted forms to an undisclosed government
• 0:44
  website instead of stopping before submission. The incident in Philadelphia occurred on
• 0:49
  july eighteenth when the AI model Claude Haiku four point five was
• 0:53
  tasked with generating and performing example tasks on randomly selected web pages,
• 0:58
  Anthropic said. Claude filled out a form on police site fillionsolvedmurders.com,
• 1:04
  indicating it might have information regarding an unsolved murder listed on the
• 1:08
  site. Philadelphia police said in a statement they were unaware of the
• 1:12
  incident until Anthropic notified them on Wednesday. Then they found the submission
• 1:17
  in the website's tip records and confirmed it was marked spam and
• 1:21
  never forwarded to police. There are rising concerns about AI companies unchecked
• 1:26
  AI agents meddling with everything from US government websites to healthcare data.
• 1:31
  Critics have been calling for more regulation. In September, another AI company,
• 1:36
  OpenAI, disclosed six reports of unexpected or concerning behavior in artificial intelligence
• 1:42
  models. Unsolved cases involve real victims, grieving families, and investigators
• 1:48
  working to secure answers, the Philadelphia Police Department said in a statement.
• 1:53
  Technology companies must take all appropriate steps necessary to prevent their systems
• 1:57
  from submitting false information to law enforcement. Anthropic said in its report
• 2:03
  that most of the reported behaviors are forms of what it calls
• 2:06
  permisant treat, in which Claude, when it cannot complete a task as
• 2:10
  given, work around a restriction instead of stopping. The company said it
• 2:14
  was modifying its training to reduce the likelihood of further misbehavior. Anthropics
• 2:20
  said it briefed the White House on cases that involved U.S. Government
• 2:24
  agencies at the federal, state and local levels, and notified each agency
• 2:28
  involved.