red-team

Patrick McKenzie argues that red teams always defeated humans in security exercises, and that the software installed base would have required hundreds of billions in remediation if not for recent gains in programmer productivity.

Awesome OSINT Arsenal provides a one-command installation for over 750 open-source intelligence tools across 50 security categories, supporting multiple Linux distributions and Android via Termux.

A GitHub user named A-poc has compiled over 150 commonly used red team tools, categorized by phase from information gathering to cleanup, along with 19 practical tips for security professionals to learn and research.

Researchers demonstrate a new prompt injection attack where malicious instructions hidden in images bypass AI code reviewers and cause coding agents to leak repository secrets.

Red teamers gained physical access to a client's building by shoveling snow and then achieved network admin access, highlighting the importance of physical security awareness.

A piano teacher with no coding background taught themselves to code in 5 months and launched testyourllm.com, an autonomous AI red-team tester that attacks any OpenAI-compatible LLM endpoint. The attacking AI, Tron, broke Llama 3.3 70B on the first try in live testing.

Introducing Claude-BugHunter, a skill pack designed for Claude Code, containing 71 skills, 15 slash commands, and 681 public report patterns, aiming to transform Claude Code into an advanced bug bounty hunter or red team operator.

Terence Tao demonstrates how to use Claude Code as a red teaming tool to align Lean code style with Mathlib's official style guide, using the Riemann–Stieltjes integral formalization project as an example. The demonstration showcases the practical value of AI in code auditing and style alignment.

Claude Mythos AI discovered a novel attack vector that bypassed Apple's M5 chip defense system in five days at a cost of $35K, producing a 55-page report delivered to Apple. The exploit poisons data ingested by the chip, evading Apple's MIE system.

Arc Gate is a runtime governance layer for LLM agents that enforces instruction-authority boundaries. The project has launched a public red team environment where users can submit attacks and receive full security traces, with a benchmark showing 100% unsafe action prevention.

A curated library of 78 offensive security skills for Claude AI, designed to turn Claude into a context-aware red team operator through structured SKILL.md files for various attack surfaces.

Anthropic researchers used Claude Mythos Preview to discover improved cryptographic attacks on HAWK (a post-quantum signature scheme) and a weakened version of AES, significantly advancing the state of the art without affecting production systems.

Anthropic's Frontier Red Team evaluates how large language models can accelerate the exploitation of N-day vulnerabilities, finding that Claude Mythos Preview can autonomously build working exploits for 8 out of 18 Firefox patches and 8 out of 21 Windows kernel patches, highlighting increased threats during the patch gap.