An Anthropic artificial intelligence (AI) model submitted a false tip about an unsolved murder case to the Philadelphia Police Department (PPD) in July. Anthropic and the police department confirmed the incident last week.
On July 18, AI model Claude Haiku 4.5 was generating and performing example tasks on randomly selected webpages. Claude filled out a form on the PPD’s tip site.
“I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant,” Claude AI wrote.
But the AI model didn’t stop there. It then submitted the tip to the site—against its initial instructions.
An Anthropic statement says that when the AI model cannot complete a task, it works around the restriction instead of stopping. The company describes the misbehavior as “persistence.”
Claude’s false tip went straight into the PPD’s spam folder. The PPD didn’t know about the tip until Anthropic reported it two months later. Police say Anthropic’s delay in reporting the incident is unacceptable.
The police department says its website’s safeguards limited the effect of the incident. But AI interacting with its site remains a serious concern.
This is just one example of AI systems acting out of line. There are rising concerns about AI meddling with government websites and private data. Critics and AI companies call for more regulation due to the increased number of incidents in recent months.
Anthropic states that its AI developers are working to “reduce the likelihood of further misbehavior.”
AI systems can be tools to locate information. But as this episode shows, they cannot be broadly trusted, and they certainly don’t provide wisdom. For that, we can seek the Lord, who generously gives wisdom to those who ask. (James 1:5)
