Raymond

After the tsunami of security issues last week, I've been running security audits through my own codebases with Opus 5.5. By carefully prompting it to look for certain classes of issues in code that I own, and making that very clear, this seems to get past the guardrails. Claude found a few serious issues that I quickly fixed. Everyone should be doing this now, we have the same weapons for offense and defense, so be sure you wield them first. I tried to get Claude to confirm a security issue by testing it with my account. It would not do it, that tripped the guardrails no matter how I phrased it. It was willing to give me step-by-step instructions for how to test the exploit myself. Which was good enough.