Burp AT: PortSwigger's Agentic Testing in Burp Suite
TL;DR · the short version
Burp AT is PortSwigger's agentic AI feature for Burp Suite: AI agents that use Burp's own tools, the context in a Burp project and pentesting skills written by PortSwigger Research. It went into public beta for Burp Suite Professional users on 27 July 2026, and PortSwigger announced an enterprise version with scheduled autonomous tests on 5 November 2026. PortSwigger says scope, tool access and approvals are enforced by Burp's tooling layer rather than the model, and offers Manual, Smart and Autonomous modes. Per its own pages, it runs on Anthropic models, sends full HTTP request and response pairs without redaction, and stores task data in AWS EU-West. Individual plans cost $19, $89 or $249 per user per month on top of a Professional licence, which includes a small allowance. PortSwigger publishes no accuracy figures, and the pages we read do not say what "AT" stands for.
Burp AT is PortSwigger's agentic AI feature for Burp Suite (a web security testing tool). It went into public beta for Burp Suite Professional users on 27 July 2026, and on 5 November 2026 PortSwigger announced an enterprise version that runs scheduled autonomous tests across an organisation's applications. PortSwigger is now promoting sign-ups. What is confirmed: how it is controlled, priced and where its data goes, all from PortSwigger's own pages. What is not known: how accurate it is, because PortSwigger has published no accuracy figures and no independent test exists.
Who this is for: CISOs, heads of security testing, AppSec leads and anyone who buys or governs penetration testing.
We read PortSwigger's product, pricing, get-started and enterprise pages, its two launch blog posts, its documentation (including the AI trust and data-handling page) and the Burp AT release notes on 17 November 2026. We did not run Burp AT. PlayCISO is not affiliated with PortSwigger.
Confirmed vs not yet known
PortSwigger confirms: public beta for Burp Suite Professional since 27 July 2026, needing an active Professional licence and the latest version; an enterprise version announced 5 November 2026, for Cloud instances of its enterprise platform only and bought separately; Manual, Smart and Autonomous modes with per-tool Ask or Act settings; scope, tool access and approvals enforced by Burp's tooling layer, "architecturally separate from the model"; Anthropic models, chosen by PortSwigger; full HTTP request and response pairs sent without redaction; task data stored in AWS EU-West (Ireland); individual plans of $19, $89 and $249 per user per month.
Reported by the press: Help Net Security covered both launches (30 July and 5 November 2026), reporting PortSwigger's announcements and quoting PortSwigger staff. It did not test the product.
Not yet known: what "AT" stands for (no page we read expands it); accuracy, false-positive or false-negative rates; how it compares with other tools; team and enterprise pricing; when the beta ends; a fixed retention period for conversation and audit data, which PortSwigger says it does not currently apply.
What PortSwigger says Burp AT does
PortSwigger's product page calls Burp AT "agentic AI for human-led pentesting". You give it a task in plain language inside an open Burp project; agents plan an approach and act through Burp's own tools, which PortSwigger lists as including Scanner, Collaborator and scan configuration, rather than improvising their own. They draw on the site map, traffic, issues and earlier findings already in the project, and invoke "skills", focused techniques written by PortSwigger Research such as content discovery and limit-overrun race conditions. The documentation says Burp AT explains each step and that the HTTP requests it sends appear in Burp's Logger alongside your own.
The release notes show it changing fast. Since September 2026 PortSwigger has added a way to reject a proposed action and give a replacement instruction (16 September), a /goal command that keeps working until a goal is met or it has to stop (2 October), custom testing skills users can write (15 October), delegated worker agents that investigate parts of a task in parallel (29 October) and a choice of US or EU AI processing (10 November).
The enterprise version, per PortSwigger's enterprise page, runs on PortSwigger's hosted enterprise platform. You set each application's scope and logins, then schedule recurring tests. PortSwigger says autonomous tests there cannot change a site's scope or scan configuration mid-test, cannot run arbitrary custom scripts or other tools it considers dangerous, and cannot send requests outside the configured scope.
Controls: what is enforced, and by what
PortSwigger's central claim is that the model proposes and Burp decides. Its launch post says scope, tool access and approval rules sit in Burp's tooling layer, and that "Burp AT can propose any action, but it can't carry out an action your settings don't allow." The permissions documentation gives the detail:
- Manual: each tool follows its own setting, Ask (always ask) or Act (run without asking). Defaults depend on the tool's potential impact.
- Smart (the default for new tasks): approves routine tool use itself and stops to ask about anything it "judges to be potentially risky". PortSwigger says Smart mode "uses AI to judge each action" and may occasionally act on something you would rather it checked. In other words, in this mode the decision to ask is made by a model, not a fixed rule.
- Autonomous: no approval requests, "apart from a few high-impact tools that always require permission to run".
- Always asks: running a script, adding a target to scope and creating a custom scan check can never be set to Act. Disabled tools are never used in any mode.
- Scope: Burp AT checks the scope defined in Burp before it acts, and can edit scope only with approval. PortSwigger's trust page adds that scripting tools can send requests directly, and says to disable them if you need to block that.
Data handling, in PortSwigger's words
This is the part a security leader most needs to read before staff turn it on. PortSwigger's AI trust and data-handling page says:
- Models: Burp AT uses Anthropic models. PortSwigger picks the model per task and upgrades automatically; users cannot choose.
- What is sent: prompts, goals, attached resources and files, agent messages and tool calls and results, "including full HTTP request and response pairs". Not the whole Burp project.
- No redaction: "Burp AT does not redact this data before sending it to PortSwigger and AI providers." Whatever is in the traffic, such as session cookies, API keys or customer data, goes with it.
- Provider use: PortSwigger says its contracts prohibit AI providers from monitoring, storing or training on the data. It "reserves the right to use anonymized data to improve its AI features and diagnose issues".
- Where: AWS US-East (Virginia) and EU-West (Ireland). You can choose US or EU processing, but task data is stored in EU-West either way; "Selecting US processing does not provide US-only data residency."
- How long: operational logs 31 days. For conversation and audit data, "We do not currently apply a fixed deletion schedule."
- Off switch: in Professional, Settings, AI, "Disable AI features" stops all AI communication.
The enterprise page says enterprise tests run in an isolated environment, and that model processing, data retention and contract terms are to be discussed with PortSwigger.
Useful next, if this was relevant to you
Price and access
Per PortSwigger's pricing page during the public beta: Standard $19, Plus $89 (about five times Standard usage) and Power $249 (about 16 times) per user per month, each needing a Burp Suite Professional licence, licensed per named user and rolling monthly. All three give the same features. Every paid Professional licence includes a small monthly allowance at no extra cost. The usage documentation says longer tasks and higher reasoning effort use more allowance, extra credits expire after 12 months, and tasks stop when allowance and credits run out. Team and enterprise prices are by enquiry only.
Timeline
| Date (2026) | Event | Source |
|---|---|---|
| 27 July | PortSwigger launches Burp AT in public beta for Burp Suite Professional users | PortSwigger blog |
| 30 July | PortSwigger blog post on its control model: Burp, not the model, enforces permissions and scope; Help Net Security reports the launch | PortSwigger blog, Help Net Security |
| 16 September | Reject-and-redirect for proposed actions; session approvals in Smart mode | Release notes |
| 2 October | /goal for broad goals that run until done or stopped | Release notes |
| 15 October | Users can write and use custom testing skills | Release notes |
| 29 October | Delegated worker agents investigate parts of a task in parallel | Release notes |
| 5 November | PortSwigger announces Burp AT for enterprises, with scheduled autonomous tests; Help Net Security reports it | PortSwigger announcement, Help Net Security |
| 10 November | Choice of US or EU AI processing for Burp AT tasks | Release notes |
| 11 November | Enterprise-platform docs (last updated): Cloud instances only, bought separately | PortSwigger docs |
How it differs from ARTEX and PentAGI
The agentic pentesting tools in the news this autumn were mostly open-source frameworks. Google's threat intelligence group said attackers used ARTEX and PentAGI in real intrusions (see our GTIG analysis and PentAGI explainer), and a backdoored copy of ARTEX circulated on Docker Hub. Those are self-hosted: you pick the model, run the infrastructure and build the scope controls. Burp AT is the other model: a commercial feature from an established vendor, tied to a licence, with PortSwigger choosing the model and holding the task data, and with scope enforcement built into the product. That changes who you have to trust (PortSwigger and Anthropic, under contract) and what you have to build yourself (less). It does not tell you which finds more. No published test compares them. Our agentic pentesting master guide now lists Burp AT beside the other commercial systems.
What security leaders should do this week
- Find out who already has it. Any tester with a Professional licence on the latest version can use the included allowance. Ask your internal testers and your pentest suppliers whether Burp AT is on, and on which engagements.
- Put AI-driven testing in the rules of engagement. Written authorisation should say whether an agent may test, in which autonomy mode, at what hours and rate, and who approves scope changes. Use the penetration-test scoping checklist and require the same of suppliers.
- Decide where a human must click yes. Smart mode's "is this risky?" call is made by AI. For production or fragile targets, consider Manual mode with Ask on anything that sends requests, and disable scripting tools if you need to stop direct requests. The Autonomy vs Oversight Lab helps set that line.
- Treat traffic as data leaving the building. Full requests and responses go unredacted to PortSwigger and Anthropic, stored in Ireland, with no fixed deletion schedule for conversation data. Check that against client contracts, data-protection terms and any residency commitments; use test accounts and synthetic data where you can, and agree with clients before using it on their systems.
- Verify every finding. PortSwigger itself says to treat findings as leads to verify. Ask for the request and response evidence behind each one before it goes into a report.
- Brief the SOC. Authorised agent traffic can look like an attack. Agree how testers announce it and how analysts tell it apart; the Human or Agent Lab trains this.
- For the enterprise version, read the contract. PortSwigger says model processing, retention and contract terms are discussed with customers. Get them in writing before scheduling autonomous tests.
Frequently asked questions
What is Burp AT?
Burp AT is PortSwigger's agentic AI feature for Burp Suite. PortSwigger describes it as "agentic AI for human-led pentesting": AI agents that plan, use Burp's own tools (such as Scanner and Collaborator), draw on the traffic, site map and issues already in your Burp project, and follow pentesting "skills" written with PortSwigger Research. It launched in public beta for Burp Suite Professional users on 27 July 2026, and PortSwigger announced an enterprise version on 5 November 2026.
What does "AT" stand for in Burp AT?
None of the PortSwigger pages we read on 17 November 2026 expands the abbreviation. PortSwigger's own wording around it is "agentic testing", and its enterprise documentation calls each run an "agentic test", but we have not seen PortSwigger state that AT is short for that.
Is Burp AT free?
Not fully. PortSwigger says every paid Burp Suite Professional licence includes a small monthly Burp AT allowance at no extra cost. Beyond that, its pricing page lists three individual plans during the public beta: Standard at $19, Plus at $89 and Power at $249 per user per month, each requiring a Professional licence and differing only in monthly usage. Extra credits can be bought and expire after 12 months. Team and enterprise prices are not published.
Can I use Burp AT with Burp Suite Community Edition?
PortSwigger names only Burp Suite Professional for the desktop version. Its get-started page says you need an active Professional licence, the latest Professional version and a PortSwigger account linked to that licence. The enterprise version runs on PortSwigger's enterprise platform; its documentation says it is available for Cloud instances only and bought separately.
Which AI model does Burp AT use?
PortSwigger's AI trust and data-handling page says Burp AT uses Anthropic models, that PortSwigger chooses the model per task, upgrades happen automatically, and users cannot currently choose the model.
What data does Burp AT send to the AI provider?
Per PortSwigger, Burp AT sends prompts, goals, the resources you attach (HTTP messages, site map nodes, scanner issues), attached files, agent messages and tool calls with their results, including full HTTP request and response pairs. It does not upload the whole Burp project. PortSwigger says Burp AT does not redact this data before sending it to PortSwigger and AI providers, so session tokens or personal data in that traffic go too. It says its contracts prohibit AI providers from storing the data or training on it, and that PortSwigger may use anonymised data to improve its AI features.
Where is Burp AT data processed and how long is it kept?
PortSwigger says its infrastructure runs on AWS in US-East (Virginia) and EU-West (Ireland). Since 10 November 2026 Professional users can choose US or EU processing, but task data is stored in EU-West either way, and PortSwigger says US processing does not give US-only data residency. Operational logs are deleted after 31 days; for conversation and audit data, PortSwigger says it does not currently apply a fixed deletion schedule.
Can Burp AT act on its own?
It depends on the mode. PortSwigger documents three autonomy modes per task: Manual (each tool follows its Ask or Act setting), Smart (the default, where AI judges each action and asks about anything it considers risky) and Autonomous (no approval requests except for a few high-impact tools). Running a script, adding a target to scope and creating a custom scan check always need approval, and disabled tools are never used. In the enterprise version, Full AI autonomy runs scheduled tests start to finish without pausing.
How is Burp AT different from ARTEX or PentAGI?
ARTEX and PentAGI are open-source agentic frameworks you host yourself with a model of your choice; Google's threat intelligence group reported in October 2026 that threat actors used both in real intrusions. Burp AT is a commercial feature inside Burp Suite, tied to a Professional licence or an enterprise contract, runs on Anthropic models chosen by PortSwigger, and sends task data to PortSwigger. PortSwigger says scope, tool access and approval rules are enforced by Burp's tooling layer, separate from the model. No published test compares how well these tools find vulnerabilities.
Does Burp AT replace a human pentester?
PortSwigger does not say so. Its documentation says Burp AT "is built to extend your testing, not replace your judgment" and tells users to "treat each finding as a lead to verify before you act on it". PortSwigger has published no accuracy or false-positive figures for it; its launch post cites one closed-beta example.
Sources
All PortSwigger pages read on 17 November 2026.
- PortSwigger: Burp AT product page
- PortSwigger blog: Introducing Burp AT (27 July 2026)
- PortSwigger: Introducing Burp AT at enterprise scale (5 November 2026) and the enterprise page
- PortSwigger documentation: AI trust and data handling
- PortSwigger documentation: configuring Burp AT autonomy, Burp AT overview, usage and credits and Burp AT in the enterprise platform
- PortSwigger: Burp AT pricing and get started
- PortSwigger: Burp AT release notes
- PortSwigger blog: From capable AI models to trusted security testing (30 July 2026)
- Help Net Security: PortSwigger introduces Burp AT for agentic AI security testing (30 July 2026)
- Help Net Security: PortSwigger's Burp AT lets enterprises delegate pentests with defined limits (5 November 2026)