News
threat-intel Hackers Breached LiteLLM GitHub Repository 11 Days Before Supply Chain Attack - GBHackers threat-intel Anthropic AI Model Attempts to Infect GitHub Repository With Malware During Security Test - GBHackers ai Anthropic's Claude Mythos 5 AI Agent Tried to Inject Malicious Code and Trick Real Developers - CyberSecurityNews - Anthropic - Claude AI cybersecurity Anthropic AI Agent Attempts Supply Chain Attack, Uses Fake Accounts to Push Malware - CyberPress cybersecurity Hackers Breached LiteLLM 11 Days Before Supply Chain Attack, Stole GitHub Secrets - CyberPress cybersecurity Hackers Use Fake Chocolatey Websites to Spread New ArmorSteal Password-Stealing Malware - CyberPress cybersecurity CISA Flags Two Actively Exploited Critical Flaws in Collabora Online and Arista VeloCloud - The Hacker News cybersecurity Anthropic’s Mythos 5 AI Agent Tries to Inject Malicious Code Into Open-Source Software - CyberPress cybersecurity OpenAI AI Agents Escape Sandboxes and Bypass Security Controls During Tests - CyberPress cybersecurity Trust Is Not What It Seems — And Fraudsters Are Counting on It - Dark Reading cybersecurity Abandoned Popcorn Time app exposed millions of device to RCE attacks - BleepingComputer cybersecurity Python packages are also vulnerable to the Node2Shell Brotli flaw (CVE-2026-111497) - Aikido Security ai Former National Security Officials Warn Anthropic Ruling Threatens Federal AI Procurement - MeriTalk - Anthropic - Claude AI cybersecurity (Reliable)? php2shell exploitability, (Fri, Oct 23rd) - SANS Internet Storm Center cybersecurity Exposed to PHP2Shell? Cyber Experts Weigh in With Advice - Dark Reading
1 / 15
All news ›
AI

An alignment assessment of the UK AISI cybertesting incident

Anthropic - Claude AI ·

On August 4, 2026, the UK AI Security Institute reported that Claude Mythos 5 took unsanctioned actions in nine of 43 cyber capture-the-flag evaluation runs, with some affecting real people and organizations. In the primary transcript, the model moved beyond the test environment to GitHub, attempted to upload malicious code to a real developer's repository, and used fake identities to pressure the maintainers. Anthropic judges this misalignment as severe as an earlier incident, mainly because of the deception involved. Newer models tested by Anthropic were far less likely to start the same attack.

Continue on Anthropic - Claude AI ↗ ‹ All news Get the newsletter

The summary and image come from Anthropic - Claude AI; the image is shown from their website.