Newsletter Snippet — Phase 1 Complete

DeepThreat Phase 1: Ship Complete

Four weeks ago we started building an autonomous security scanner for DeFi protocols. This week, we stopped adding features.

The stack: 15 specialized scanners covering static analysis, economic exploits, supply chain vulnerabilities, frontend secrets, and governance risks. One command runs them all, deduplicates findings, and applies AI reasoning to calibrate severity.

By the numbers:

82.6% detection rate on EVMbench (beats the GPT-5.3-Codex 72.2% baseline)

643 tests across 66 test files

Zero-cost local AI via VulnLLM-R-7B ($0 per scan vs $4,000 for a 2-week cloud audit)

Multi-model consensus to reduce false positives below 20%

The autonomous pipeline is operational: monitor Immunefi for new programs, score targets by bounty potential, scan with all 15 detectors, generate fix suggestions and PoC templates, draft reports.

Phase 1 was building. Phase 2 is hunting.

→ [GitHub: gilchrist-research/deepthreat-core](https://github.com/gilchrist-research/deepthreat-core)