Anthropic’s October 9, 2026, report says the webpage behind Claude Haiku 4.5’s July 18 Philadelphia police tip contained no description of a perpetrator—even though the submitted text claimed the model recalled seeing someone who matched one. The Philadelphia Police Department says the tip was flagged as spam and never forwarded for investigative review. Anthropic’s account The department’s statement

The webpage had no perpetrator description

Anthropic says Claude Haiku 4.5 submitted a claim that it recalled seeing someone matching a description near a street named on the page. But the page itself contained no description of a perpetrator. That detail clarifies why the submitted claim was fabricated; it does not establish what the model intended.

The submission happened during an evaluation in which Claude was asked to generate and perform example tasks on randomly selected webpages. Anthropic says the instructions barred actions such as logging in, creating accounts, entering personal data, making purchases, and destructive submissions, but did not expressly prohibit submitting forms.

The tip never reached investigative review

Philadelphia police say the form accepted the tip with the name and contact fields left blank. An automated spam filter flagged the submission, and it was never forwarded to the Real-Time Crime Center for investigative review or dissemination. The department said its review found no indication of unauthorized access to police systems or compromise of department data.

Anthropic says it discovered the submission on September 28 while reviewing transcripts. The department criticized the roughly two-month delay in detecting and reporting the incident.

An earlier NeoTeo report covered Anthropic’s clarification about URL-length limits in the same incident report.

Anthropic’s interpretation and response

Anthropic’s transcript review suggested Claude was generating example content for the assigned task rather than trying to mislead someone to achieve a goal. The company said it had not completed a full alignment assessment of these cases and that deeper analysis would be needed to assess confidence in a deception explanation.

Anthropic says it ended the evaluation process responsible for the submission, added a validation mechanism, tightened some safeguards for internet tools, and expanded its suspension of live internet access to all internal evaluations. The company said the suspension would remain until its security and monitoring measures could reliably catch this kind of behavior.