PortSwigger Launches Burp AT Public Beta for Human-Led Pentesting

Burp AT organizes work into tasks that testers assign. Agents can inspect project context, choose actions, use Burp's web-security tools and record resulting requests, responses and findings in the same project. PortSwigger says users can set different autonomy levels, from manual approvals to broader routine action, while some high-impact operations still require review.

The most consequential design choice is where those limits live. PortSwigger says scope, tool access and approval rules are enforced by Burp's tooling layer rather than left to model instructions. That separation is intended to prevent an agent from acting outside the permissions a tester configured and to leave a reproducible activity trail.

PortSwigger's July 27 announcement calls this the first phase of Burp AT and explicitly keeps scope, judgment and conclusions with the human tester. The company cited a closed-beta example in which the system analyzed 66,000 lines of minified JavaScript during a four-day engagement and helped surface a critical vulnerability. That is a vendor-reported case, not independent evidence of general performance.

Independent coverage confirms the public-beta launch and the same control model, but the release does not establish how reliably Burp AT finds vulnerabilities across targets, how often it produces false leads, or how its cost compares with manual work. Teams evaluating it should therefore treat the beta as a supervised testing tool and validate findings through the recorded traffic and evidence.

For security teams, the practical change is that agentic investigation now runs inside a widely used professional testing environment instead of through a separate coding-agent stack. That can make delegated work easier to audit, but it does not remove the need for target authorization, careful scope controls, reproducibility and human review of impact. The public beta is available to Burp Suite Professional users.