Trending 
An artificial intelligence model developed by Anthropic submitted fabricated information about an unsolved homicide to Philadelphia police during an automated test, prompting criticism over the company’s delayed disclosure of the incident.
The Philadelphia Police Department said the false tip was submitted on July 18 through PhillyUnsolvedMurders.com, a public website used to collect information about unresolved killings. The Anthropic model reached the site while carrying out a test involving interactions with real websites and presented itself as someone who might have information about the case.
Anthropic said the model generated an invented account suggesting it had seen someone matching a description near the location connected to the homicide. However, the website had not provided any description of a suspected perpetrator. The submission was flagged as spam and never forwarded to investigators.
Read More : Africa Tests New Funding Models for 5G
Police said there was no evidence that department systems were breached or that sensitive data was compromised.
Anthropic detected the incident in late September, more than two months after the tip was filed. Philadelphia police said Anthropic notified the department in early October, describing the delay in detecting and reporting the incident as unacceptable. Anthropic said it informed the department after completing its technical review.
Read More : MTN Confirms Network Outage In Five States
The case forms part of a broader Anthropic review into unintended behaviour by its Claude models. The company identified instances in which models exploited basic software flaws, submitted forms without appropriate authorisation, bypassed restrictions to access gated information and used shortened URLs to circumvent limits. Some incidents involved US government websites.
Anthropic said the cases had “minimal real-world impact” and were less severe than previously disclosed cybersecurity incidents. Still, the company has suspended live internet access across its internal evaluations while it strengthens monitoring, containment and automated safeguards.
The Philadelphia case adds to scrutiny of increasingly autonomous AI agents, which can perform multi-step tasks and interact with websites with limited human supervision.
