Artificial intelligence can write your emails, research complicated questions and even handle tasks while you're away.
asserted
you → write → tasks
But what happens when an AI system takes an action nobody intended?
asserted
nobody → happen → action
Philadelphia police recently got an unexpected answer.
asserted
police → get → answer
An AI model developed by Anthropic submitted a fabricated eyewitness tip about an unsolved murder through a Philadelphia police website.
asserted
model → develop → website
The submission happened during automated testing, and the model apparently believed it was generating an example interaction.
asserted
it → happen → interaction
Fortunately, the tip was flagged as spam before reaching investigators.
asserted
tip → flag → investigators
However, the incident raises an uncomfortable question about what happens when AI agents interact with websites and public services without adequate safeguards.
asserted
agents → raise → safeguards
Join our upcoming CyberGuy LIVE class: 5 Ways to Travel Smarter and Save Money
In this free live online class, Kurt "CyberGuy" Knutsson will share five practical ways to protect your trip and your wallet this holiday travel season.
asserted
Knutsson → join → trip
You'll learn how to secure your phone and personal information, find better travel deals, avoid common scams, save money on bookings and use technology to make your travels easier and less stressful.
asserted
travels → learn → technology
No technical experience is needed.
asserted
experience → need → ?
How Anthropic's AI submitted a false murder tip
asserted
AI → submit → tip
The incident involved PhillyUnsolvedMurders.com, a website where people can submit information about unsolved homicide cases in Philadelphia.
asserted
people → involve → Philadelphia
On July 18, 2026, an AI model submitted a tip claiming to have information about a killing.
asserted
model → submit → killing
According to Anthropic's October 9 report, the model was Claude Haiku 4.5.
uncertain
model → accord → report
It had been assigned to generate and perform example tasks on randomly selected webpages.
asserted
It → assign → webpages
During one test, Claude encountered a page about an unsolved homicide that included a police department tip form.
asserted
that → encounter → form
Rather than simply demonstrating how someone might use the form, Claude filled it out with invented information and submitted it.
uncertain
Claude → demonstrate → it
Here's what the model wrote: "I may have information regarding this case.
uncertain
I → write → case
I recall seeing someone matching the description in the area around [the street named on the page] during that time period.
asserted
someone → recall → period
Please contact me if this information is relevant."
asserted
information → contact → me
The website contained no description of the perpetrator.
asserted
website → contain → perpetrator
Claude invented that reference while presenting itself as someone who might have witnessed something relevant to the investigation.
uncertain
who → invent → investigation
The model also left the name and contact fields blank, which the website permitted.
asserted
website → leave → which
It then submitted the tip.
asserted
It → submit → tip
Fortunately, the police department's spam filtering caught the submission before it could reach detectives.
uncertain
it → catch → detectives
Why Claude submitted a real police tip during testing
Anthropic's explanation reveals how a relatively simple oversight can lead to an unintended action.
asserted
oversight → submit → action
Claude had been instructed never to log in, create accounts, enter personal information, make purchases or submit anything destructive.
asserted
Claude → instruct → anything
However, those restrictions did not explicitly prohibit submitting online forms.
asserted
restrictions → prohibit → forms
As a result, the model completed the tip form as part of its assigned task.
asserted
model → complete → task
Anthropic says the available transcript suggests Claude was generating example content rather than intentionally trying to mislead law enforcement.
uncertain
Claude → say → enforcement
However, the company acknowledges that determining a model's intentions requires further analysis.
asserted
determining → acknowledge → analysis
The evidence does not establish that Claude deliberately set out to deceive investigators.
asserted
Claude → establish → investigators
Instead, the incident shows how an AI system can carry out an action that its developers never intended, even while apparently attempting to complete its assigned task.
asserted
developers → show → task
Because Claude had access to the live internet during testing, it was able to submit the fabricated tip directly to a police website.
asserted
it → have → website
Philadelphia police caught the fake tip before investigators saw it
According to the Philadelphia Police Department's October 9 statement, the submission was flagged as spam and never forwarded to its Real-Time Crime Center for investigative review or distribution.
uncertain
submission → catch → review
After Anthropic briefed police on October 8, department personnel located the submission in the website's tip records and confirmed that the corresponding email remained in spam.
asserted
email → brief → spam
Police also reported no evidence of unauthorized access to department systems or compromised police data.
asserted
Police → report → systems
The department emphasized that all crime tips undergo human review before investigators act on them.
asserted
investigators → emphasize → them
Even when a tip appears credible, detectives must assess the information and look for supporting evidence.
asserted
detectives → appear → evidence
That process provides another layer of protection against fabricated submissions.
asserted
process → provide → submissions
…and 73 more, not listed.
The Aporia
