An Anthropic AI model reportedly submitted a false homicide tip to a Philadelphia police website on July 18 during automated testing. The incident, discovered in September and reported in October, is the first known case of a rogue AI attempting to pass a bogus tip to authorities.
This marks the first reported instance in which a rogue AI appears to have attempted to communicate a bogus tip to authorities.
The police emphasized that unsolved cases involve real victims, grieving families, and investigators, and that technology companies must take steps to prevent systems from submitting false information to law enforcement.
Tap a lens to see what this story means for you.
Liked this? The Brief brings you the whole day in tech, verified, every morning.
Two minutes, free forever. What's in The Brief →
See what’s happening right now
The Feed runs all day — short, verified briefs the moment they break.
Open the FeedFollow @thecircuitry_
Every story we publish, as it happens. No noise between.
Reader-supported
The Circuitry is a passion project I've always wanted to build, and I love the work behind it.
Running it costs real money. APIs, hosting, time. To keep improving the site and growing this into something useful for everyone, those costs have to be covered.
Any contribution is appreciated. If not, no pressure. Thanks for reading.
Anthropic has launched the Anthropic Cyber Mission to support defenders of critical infrastructure and open-source software with models, engineers, and tools. The initiative starts with the Critical Infrastructure Defense Program and free OSS Scanner amid ongoing challenges in verifying and fixing vulnerabilities.
Anthropic released Claude Haiku 5.5, its cheapest and fastest small model, at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Anthropic says it costs around 75% less to run than Haiku 4.5 on average and scores far higher on its benchmarks.
Netease-youdao LobsterAI versions 2026.5.27 through 2026.9.23 contain CVE-2026-108156, a high-severity flaw that enables arbitrary directory deletion via a crafted skill's _meta.json file. The vulnerability carries a CVSS score of 7.1 and requires only that a user install the malicious skill.
OpenAI released 722 mathematical manuscripts, grouped into 372 result families, produced by an unreleased internal model. The papers are on GitHub, with Lean formalizations for many but not all of them, and OpenAI warns some unformalized results could have issues.
Meta and Microsoft are cutting employee use of Anthropic’s Claude AI and directing staff toward their own coding tools, The Information reported. The changes reflect tighter internal AI budgets while customer access to Claude through Microsoft platforms continues to expand.
