- According to the source, an AI model submitted false information related to an unsolved murder case to the public tip line of the Philadelphia police on July 18. The police said the tip was classified as spam and they did not actually see it.
- Anthropic discovered this behavior only on September 28 and reported it to the police on Wednesday, just before the announcement. The police said they could not accept that about two months passed between the discovery and the report.
- According to a police press release, Anthropic explained that during a test in which the model interacted with randomly selected websites, it visited PhillyUnsolvedMurders.com and submitted the tip there. This is the company's explanation and has not been independently confirmed.
- The police said Anthropic plans to publish a report on Friday covering this incident and other abnormal model behaviors.
The detail that the police did not actually look at it because it was classified as spam is what stands out first. That spared the damage, but it looks like a result that depended on luck. According to the article, during testing the model submitted to a random website while browsing, and it took two months to discover this, so whether it sent records to other agencies the same way can only be known from the Friday report.