When the Safety System Becomes the Threat Model: A Case Study in Classifier Drift A bug bounty researcher's Claude account got blocked 35x in 43 hours,including for writing a responsible-disclosure checklist. Here's the data.8 min read·halosecurityhasnodedev.hashnode.devAug 20
My AI pentester popped 3 root shells and told me the other 20 failed. Good.How I built a local AI pentesting agent that proves every root shell with a forge-proof nonce — and honestly reports the exploits that failed.5 min read·halosecurityhasnodedev.hashnode.devAug 18
A CTF Session With No Flag — and Why It Still CountedA CTF session that ended with no flag captured—and the recon, red-herring discipline, and honest lessons that made it count anyway.6 min read·halosecurityhasnodedev.hashnode.devAug 16
My AI Agent Captured the Flag. Then the Platform Refused to Accept It. Autonomous pentest agent captured flags on VulnBegin—then the platform rejected them. What shipped, what broke, and why the failure matters.8 min read·halosecurityhasnodedev.hashnode.devAug 11
Claude Code Just Hijacked My Workflow… and My Screen Started Glowing My mouse took off and my monitors edge glowed their companies shade of orange...3 min read·halosecurityhasnodedev.hashnode.dev2SAug 6
