Using TypeSafe AI for High-Speed Bug Bounty Reconnaissance

Summary

A security researcher describes using TypeSafe, an AI model designed for fast categorization and decision-making rather than conversational interaction, in a bug bounty reconnaissance workflow. The service accepts multiple-choice, scoring, and true-or-false questions, returning answers with confidence or probability scores; the author cites a price of $0.042 per million input tokens and free output tokens. In an initial test, it processed 17 subdomains per second, or about 1,040 domains per minute. The workflow applied broad questions to webpage titles, headers, and visible HTML to identify portals, login pages, directory listings, buckets, outdated installations, internal tools, security products, exposed sensitive data, and hosts worth further investigation. Because the system analyzed HTML instead of screenshots, the author says it could also detect data such as API keys directly present on a landing page. Results were stored in CSV files, then Claude selected high-scoring results for text-based review, while the author used a browser extension for manual inspection before sending interesting cases to Claude for additional analysis. The article presents the tool, also referred to as Jev, as a fast filtering layer rather than a replacement for manual testing or more capable models. Its purpose is to reduce the number of hosts sent to expensive frontier models and focus human and compute resources on the most promising targets. The author describes the work as an early, first-day experiment and points readers to a related GitHub project.