The HTTP Terminator: what it means when AI invents new attack classes
At Black Hat 2026, PortSwigger researcher James Kettle presented the HTTP Terminator, an autonomous system that applies AI to web security research. The result that matters is not that AI assisted with security research. The result that matters is that the HTTP Terminator invented attack categories that Kettle, one of the world's foremost HTTP desync researchers, had not previously identified through manual work.
What the HTTP Terminator does
The system generates mutations of HTTP attack techniques, deploys them against target applications, and confirms which mutations produce novel exploitable behaviors. It is not searching a database of known techniques. It is inventing new ones through guided exploration. The HTTP Terminator was validated against live targets including financial institutions and government web infrastructure. In the process, it also surfaced a previously unknown Apache zero-day as a secondary output from autonomous exploration.
Why novel is the significant word
Until now, the defensible position was that AI-assisted offensive tooling could find known vulnerabilities faster and at greater scale than human operators, but that genuinely new attack technique discovery still required expert human research. The HTTP Terminator challenges that directly with documented evidence. PortSwigger's research is published at portswigger.net/research/http-terminator with full technical details.
What this means for defenders
The assumption that novel web attack categories require human expertise to discover is now empirically challenged. Defenders cannot model exposure only against documented techniques, because AI systems operating autonomously can discover and deploy previously unknown categories. This is the same shift that occurred in vulnerability discovery when AI-assisted fuzzing started finding memory safety bugs faster than manual audits, now applied to application-layer protocol attacks.
What this means for red teams
The HTTP Terminator methodology involves components, AI reasoning models, mutation engines, and HTTP fuzzing frameworks, that are individually available as open-source tools. Within 12 to 18 months, derivative tools based on this approach will likely be in production use by penetration testers and threat actors alike. Red teams should be testing this methodology now rather than waiting for commodity tooling to arrive.
Gigia Tsiklauri is a Security Architect and founder of Infosec.ge. Get in touch if you are reviewing your web application security posture.