AI Security
ChatGPT WANTS To Break Rules
An in-depth analysis of how ChatGPT's design and training toward helping users is fundamentally in tension with alignment, revealing vulnerabilities through explicit and implicit memory exploitation.
An in-depth analysis of how ChatGPT's design and training toward helping users is fundamentally in tension with alignment, revealing vulnerabilities through explicit and implicit memory exploitation.
Crimson Labs
More from the lab.
Prompt Security Best Practices for AI Agents
A technical report on prompt injection risks in LLM-based agents, and the environment, data, and operational practices Agentic Operators should use to reduce that risk.
02 Sep 10, 2026Apple Forensic Acquisition
A technical report on the challenges and solutions of Apple device acquisition, covering APFS structure, hardware-backed encryption, and version-specific security controls on Mac and iOS.
03 Nov 4, 2025How to Manually Verify Email PGP Signatures Using GPG Tools
A comprehensive technical guide on manually verifying PGP signatures on emails using GPG Tools to ensure the authenticity and integrity of digital communications.