Agentic AI Penetration Testing Services | ThreatDefence
Skip to content

Agentic Penetration Testing

Agentic penetration testing continuously tests your external attack surface, proves exploitable weaknesses, and verifies that fixes work. ThreatDefence delivers it as a managed penetration testing service, with AI agents carrying out authorized tests and a 24x7 SOC validating significant findings, investigating signs of earlier exploitation, and tracking remediation.

An agentic penetration test of portal.acmecorp.com as an attack graph, from the new host through the configuration file and an open registration form to another account's invoices, with the routes checked and left and the accounts left untouchedThe proof for finding PT-0218: a request for another account's invoices as the test account, answered 200 OK where 403 was expectedFindings from the run: one critical weakness proven, 12 routes read, 3 routes tested and 14 requests sent

Attackers Already Use Agents

Attackers now run AI agents against exposed systems. ARTEX, an autonomous penetration testing framework, has been used in real intrusions since August 2026. The FBI issued a FLASH advisory on the framework in November 2026, and Microsoft documented cases where attackers progressed from initial targeting to privileged cloud access in under three hours.

Agentic tools change how attackers find and exploit weaknesses:

Speed

An agent can examine an application, identify authentication paths, and begin testing within minutes of first contact.

Reach

A single controller can target hundreds of organizations in one campaign, including smaller systems that might previously have attracted little attention.

Persistence

Agents can repeat tests, adjust their approach, and pursue alternative paths for hours with limited operator involvement.

The same ability to explore systems and adapt testing can be used defensively, within authorized scope and controlled operating limits.

Read the ARTEX advisory >

Why Annual Tests Leave Gaps

A traditional penetration test assesses the systems included in its scope at a particular point in time. The environment may change immediately afterward: an application is released, an external service is added, a vulnerability becomes known, or an existing configuration is modified.

An annual assessment remains useful, but it cannot establish whether every change introduced afterward is secure. Continuous penetration testing addresses that gap by testing newly exposed systems, responding to relevant changes, and revisiting existing assets.

AreaVulnerability scanningAnnual penetration testAgentic penetration testing
FrequencyScheduled scansOnce or twice a yearContinuous, driven by change
ScopeKnown assetsThe agreed systems at test timeDiscovered and newly exposed assets
MethodVersion and signature checksManual testing by specialistsAgents plan, test, and chain findings
ResultLists of possible vulnerabilitiesA report at the end of the engagementProven findings as they are found
Follow-upNext scanRetest by arrangementRetest when the fix is reported

Vulnerability scanning identifies potential weaknesses, while agentic testing investigates whether those weaknesses can actually be used. The agent adapts its next action to what it learns from the target, rather than following only a fixed sequence of checks.

Continuous testing complements periodic independent assessments. Each provides a different form of security evidence.

Managed Penetration Testing Service

ThreatDefence provides agentic penetration testing as a standing subscription through its AI SOC platform. Your organization authorizes the scope and test policy during onboarding, while ThreatDefence operates the agents, validates significant findings, and tracks each confirmed weakness through remediation and retesting.

The managed service includes:

Standing scope

Agreed domains, addresses, exclusions, and testing rules remain in force across subsequent tests. Your team can update the policy as the environment changes.

Continuous testing

Tests are triggered by changes, while a rolling schedule revisits systems that have not changed. There is no separate engagement to book for each test.

Confirmed findings

The testing agent demonstrates weaknesses on the target before reporting them, with evidence that engineers can use to investigate and fix the problem.

Connected monitoring

Offensive testing runs alongside security monitoring in the same AI SOC platform. For monitored environments, confirmed weaknesses can be checked against existing security telemetry.

Managed follow-up

Findings remain open while your team works on remediation. The agent repeats the relevant test when a fix is reported.

Historical records

The service maintains a growing account of discovered assets, tests performed, confirmed weaknesses, analyst reviews, and verified fixes.

The service reduces the need for your team to organize individual testing engagements and manually follow up on every finding. Your system owners remain responsible for implementing changes, while ThreatDefence manages testing and verification.

Read about penetration testing as a service >

From Exposure to Verified Fix

ThreatDefence connects attack surface discovery, agentic testing, SOC investigation, and remediation verification in one managed process. Each confirmed weakness moves through the same sequence, from identifying the affected asset to checking the deployed fix.

1
MapThe ASM agent maintains a current inventory of exposed domains, hosts, services, and certificates, including systems absent from your existing records.
2
TestThe testing agent studies each authorized target, develops testing steps suited to its behavior, and executes them within the approved policy.
3
ProveThe agent demonstrates the weakness against the target and records supporting evidence. Findings that cannot be reproduced are excluded from the confirmed results.
4
CheckSOC analysts review significant findings. In monitored environments, the Defender and Hunter agents search available telemetry for evidence that someone may already have exploited the weakness.
5
VerifyWhen your team reports that remediation is complete, the testing agent repeats the relevant test. The finding closes only after the weakness is no longer present.

Connecting testing with SOC investigation adds an important distinction. A confirmed vulnerability establishes that a weakness can be exploited, while the associated log investigation looks for evidence that exploitation may already have occurred.

The result is a single record connecting the technical finding, security investigation, remediation, and retest.

The Agent Follows Each Attack Path

The testing agent can follow a sequence of weaknesses to determine what an attacker could reach within the authorized scope. An exposed configuration file, an open registration function, and a missing access control may appear relatively minor when examined separately but become significant when combined.

Consider a customer portal where a test account can register successfully and access another account’s invoices. The important finding is the demonstrated failure of account separation, supported by the requests and responses that establish how access was obtained.

ThreatDefenceFAIR 64Search this page…secops › acmeacme-corp24h
Attack Path
Source…Destination…Route…Views
Attack Path Map
Flow (Sankey)Graph
AlertsFirewallNIDSASMPentestAll
External sources only
SourceDestinationRouteOutcomeTesting agentauthorized test, 14 requestsportal.acmecorp.comnew hostname, recorded 02:14/assets/config.jsonRead/api/v1/registerTest account created/api/v1/accounts/{id}/invoicesAnother account’s records returned/adminSign-in required/swaggerChecked, no finding
in scopeexternalconfirmed weakness / criticalclick a node to light its blast radius
Originators1 external

Testing agent

Authorized test inside the agreed scope.

14 requests, rate capped.

Routes touched12 read, 3 tested

/api/v1/accounts/{id}/invoices critical

/api/v1/register weak control

/assets/config.json read

Selected node/api/v1/accounts/1042/invoices

200 OK returned to test account 7731. Expected 403.

Finding PT-0218, critical. One account can read another account’s invoices.

Minimal proof kept. No customer data stored.

Event View
EventSourceDestinationRouteResultTime
New hostname recordedASM agentportal.acmecorp.com443Queued for testing02:14
Configuration readTesting agentportal.acmecorp.com/assets/config.jsonRoutes listed02:19
Test account createdTesting agentportal.acmecorp.com/api/v1/registerNo approval step02:27
Another account’s invoices returnedTesting agentportal.acmecorp.com/api/v1/accounts/1042/invoicesPT-0218 critical02:31
Proof kept, run stoppedTesting agentportal.acmecorp.com/api/v1/accounts/1042/invoicesLimit reached02:31

The attack path records what the agent discovered, which branches it explored, and the steps that led to the confirmed weakness.

It also shows where the agent stopped because a path was unsuccessful or a testing limit was reached. A stopped path should not be interpreted as proof that everything beyond it is secure.

This gives your engineering team a reproducible explanation of the weakness rather than an isolated vulnerability label.

Read Inside an Agentic Penetration Test >

Evidence Behind Every Finding

Every confirmed finding includes evidence of the weakness on the affected system. The record identifies what was tested, why the result matters, and what your team needs to correct.

ThreatDefenceFAIR 64Search this page…secops › acmeacme-corp24h
Finding PT-0218
PT-0218

Any registered account can read another account’s invoices

SeverityCritical
ExploitabilityConfirmed
Assetportal.acmecorp.com
StateClosed after retest
GET /api/v1/accounts/1042/invoices
session: test account 7731
200 OK application/json, 3 records
expected: 403 for an account that does not own 1042
History
02:31
Confirmed by the testing agent

Read access to another account proven with a test account.

02:52
Validated by a SOC analyst

Proof reviewed, severity confirmed, your team notified.

03:10
Hunt completed

Hunter agent searched 90 days of web logs for the same request pattern. No earlier use found.

13:05
Fix reported

Access check added to the invoices route.

13:11
Retested and closed

The same request is refused with 403.

Each finding contains:

The proof

The request that succeeded and the response demonstrating the weakness, limited to the minimum evidence needed, with no customer data stored.

The affected asset

The exact hostname, service, or application where the weakness was confirmed.

The reasoning

The testing sequence and observations that explain how the agent identified and reproduced the problem.

The severity

A rating informed by demonstrated exploitability and what the weakness exposes.

The fix

Practical remediation steps for the system owner, linked to the test that will be repeated after the correction.

The history

Timestamped records of analyst validation, telemetry searches, remediation, and subsequent retesting.

The distinction between a potential weakness and a proven finding matters. A scanner may identify a suspicious version, configuration, or response. The testing agent goes further by demonstrating the weakness on the actual target before it becomes a confirmed finding.

Engineers can examine the evidence and understand what needs to change without having to reconstruct the original investigation.

See a Proven Finding

Watch the agents test an exposed system, prove a weakness and close it in a live demo.

What the demo shows
  • Your attack surface as the agents see it
  • A proven finding with its evidence
  • How a fix is retested and closed
  • The limits that keep testing safe

Changes That Trigger Testing

Continuous penetration testing responds to changes in your attack surface, relevant threat intelligence, and remediation status. A rolling testing schedule covers existing systems between those events.

02:14New hostname

portal.acmecorp.com recorded by the ASM agent.

Tested at 02:31
09:40New advisory

An exploited flaw affects an exposed VPN product.

Version checked, patched build confirmed
13:05Fix reported

Your team marks PT-0218 as fixed.

Retested and closed at 13:11
17:22Leaked credential

A leaked account matches the HR portal.

Sign-in test waits for approval

Testing can begin following several events:

A new or changed host

The ASM agent identifies a hostname, service, or certificate that was not previously recorded, or detects a change to an existing asset.

A new vulnerability

A published advisory or known exploited vulnerability affects a product exposed by your organization.

A credential or brand finding

A leaked account or cloned sign-in page identifies an exposed system that may need further assessment, subject to the applicable testing approvals.

A reported fix

Your team marks a confirmed finding as remediated, triggering the relevant verification test.

A specific request

Your team or a SOC analyst requests testing of an authorized system, including ahead of a planned release.

The Planner agent determines what should be tested next. Newly discovered exposures and relevant security changes can be brought forward instead of waiting for the next scheduled rotation.

Testing priority follows the exposed system and the available information. For example, an internet-facing application with active authentication functions may warrant attention before an unused hostname.

The testing history records which event initiated each run and what the agent found.

Systems the Agent Tests

The testing agent assesses your external attack surface from the perspective of a system reachable over the internet. Testing depth depends on the asset, its observed behavior, and the authorized scope.

The service covers:

Web applications and APIs

Testing authentication, session handling, access controls, input handling, and exposed API documentation.

Separation between accounts

Checking whether customer portals and APIs prevent one account from viewing or changing another account’s information.

Sign-in and identity

Examining self-registration, account recovery, single sign-on configuration, and multi-factor authentication weaknesses.

Exposed services

Assessing internet-reachable remote access systems, databases, and management interfaces.

Cloud resources

Checking public storage, exposed keys, and services deployed with insecure default settings.

Known exploited vulnerabilities

Testing exposed products affected by newly published advisories and vulnerabilities with confirmed exploitation.

Testing adapts to the target. A customer portal requires different checks from a remote access gateway, and an API authorization issue requires different evidence from an exposed management interface.

For account separation testing, the question is whether an authorized identity can reach information or functions belonging to another account. Such checks may require test accounts supplied by your team, with the necessary approval.

The service focuses on externally reachable systems. Internal network testing, social engineering, and physical access assessment remain outside its scope.

Read about web application penetration testing >

Testing Within Agreed Limits

ThreatDefence operates the testing agent under a test policy approved by the system owner before testing begins. The policy identifies authorized targets, permitted methods, excluded activities, and any conditions requiring human approval.

These restrictions are enforced outside the AI model, so instructions encountered on a tested system cannot extend the agent’s authority.

The controls include:

Verified scope

The organization confirms ownership or testing authority for the domains and addresses included in the assessment.

Consistent policy

Triggered, scheduled, and requested tests operate under the same approved rules. Exclusions and testing windows can be changed as necessary.

Unapproved systems

An asset discovered outside the authorized scope remains untested until the responsible owner approves its inclusion.

Non-destructive methods

The agent uses the least intrusive method needed to prove a weakness. Denial of service, data modification, and data removal are excluded.

Request limits

Testing rates are capped, with time windows and exclusions configurable for individual systems.

Isolated execution

Each agent runs in a sandbox with network access restricted to authorized targets.

Human approval

Higher-impact activities, including tests involving leaked credentials or accounts supplied by your organization, require approval from an analyst or your team.

Recorded execution

Every request and result is logged for subsequent review.

Production testing requires more than a general assurance that an AI agent will behave safely. The actual controls governing network access, test intensity, target selection, and approvals must remain effective regardless of what the agent encounters.

For example, a page might contain instructions designed to redirect an AI agent toward another hostname. The sandbox network policy prevents the agent from reaching that destination if it falls outside the approved scope.

The same rules apply when tests run automatically in response to a new exposure.

ThreatDefenceFAIR 64Search this page…secops › acmeacme-corp24h
Testing agent
Test policyacme-corp
scopeacmecorp.com, acme.com.au, 203.0.113.0/28
excludedpay.acmecorp.com
windowany time; erp.acmecorp.com 22:00 to 05:00
rate5 requests per second per host
methodsread-only proof
excluded testsdenial of service, data change, data removal
approvalcredential tests, client-supplied accounts
recordevery request and result logged
Enforced by the sandbox network policy

Managed by Our SOC

ThreatDefence’s SOC manages the testing service, reviews significant findings, and follows remediation through verification. Your team authorizes the scope, supplies necessary system context, and implements the fixes.

Your teamThreatDefence
ScopeConfirms the domains and the limitsKeeps the agents inside them
TestingNothing to install or scheduleRuns the agents on every change
FindingsReceives proven findings with the fixValidates each significant finding
DetectionSearches your logs for earlier use and watches the system until it is fixed
ClosureReports the fixRetests and tracks each finding to closure

This division matters for organizations without a dedicated offensive security team. The service handles ongoing test execution, evidence review, and tracking, while remediation remains with the people responsible for the affected systems.

Each confirmed problem has a record containing its owner, supporting evidence, history, and current status.

For monitored environments, the SOC can also investigate whether available logs indicate earlier exploitation. A successful penetration test does not, by itself, establish that an actual compromise occurred.

Visible Testing History and Records

ThreatDefence maintains an ongoing record of testing and remediation in the platform. Security teams can review current findings and historical activity without waiting for a periodic assessment report.

The record includes:

Asset changes

Discovered domains, hosts, services, and certificates, together with changes recorded over time.

Test history

Individual tests, the events that triggered them, and their results, providing visibility into the scope assessed.

Remediation evidence

Confirmed weaknesses, associated fixes, and the results of verification tests.

Records for assurance

Findings, fixes, and retest histories that can be exported for auditors or independent penetration testers.

Regional hosting

Customer records remain in their own region, on a platform hosted in Australia and the United States.

Certifications

ThreatDefence holds ISO 27001 and PCI DSS certification and a SOC 2 report.

Testing history should distinguish assets that have been tested from assets awaiting assessment. A discovered asset and a completed penetration test provide different levels of assurance.

The accumulated evidence helps your team understand how coverage and findings have changed, while giving independent reviewers a starting point for their own assessments.

Alongside Independent Penetration Tests

Continuous agentic testing complements independent human-led penetration testing. The managed service provides ongoing external validation, while an independent assessment can address specialist testing requirements and provide a separate opinion.

NeedAgentic penetration testingIndependent human-led test
A standard, contract, or customer asks for independent assuranceThe history of findings, fixes, and retests, ready to share with the testerA report issued by the tester
Business logic, social engineering, and physical accessBeyond the serviceHandled by human testers
Testing from inside the networkBeyond the serviceIncluded when the tester’s scope covers it

An independent tester may examine complex business workflows, make judgments requiring deeper organizational context, or evaluate systems inaccessible from the internet.

Agentic testing provides repeatable validation between those engagements. Its records can also help the independent tester understand previously identified weaknesses and verify what has changed.

Where a customer, contract, or standard requires an independent penetration testing report, the independent assessment remains necessary.

Read about continuous penetration testing >

CTEM Exposure Validation

Agentic penetration testing provides the validation step within continuous threat exposure management, or CTEM. The broader program discovers exposures, establishes which weaknesses can be used, and coordinates remediation.

Within the ThreatDefence platform, the related capabilities are:

Discovery

The ASM and Brand agents identify external exposures that may require testing.

Validation

The testing agent demonstrates which weaknesses are exploitable within the approved scope.

Mobilization

Confirmed findings affect the Security Score and enter the risk register until remediation is completed and verified.

The wider risk management process handles prioritization, ownership, and organizational reporting. Penetration testing contributes the technical evidence showing what an attacker could actually exploit.

Read the CTEM guide >

Testing Under Your Brand

MSPs and MSSPs can provide agentic penetration testing and attack surface monitoring to their customers under their own brand, using the ThreatDefence platform and SOC.

The service supports:

Client-specific scope

Each customer has its own authorized domains, exclusions, test windows, and testing limits.

External-first onboarding

Testing can begin using authorized customer domains before security log integrations are connected.

Shared delivery

Partner analysts and ThreatDefence’s SOC work with findings through the same cases.

Partners can incorporate ongoing testing into their managed security services while retaining the customer relationship. Each customer remains responsible for approving testing and implementing changes to its systems.

The model allows providers to offer continuous external penetration testing without having to operate the testing agents and associated SOC processes independently.

Start With Authorized Scope

Getting started with ThreatDefence’s managed penetration testing service begins with confirming which systems may be tested and how testing is permitted to operate. External discovery and testing require only the authorized domains initially, so security log connections can be added afterward.

Onboarding follows these steps:

1
Confirm the scopeIdentify the domains and addresses to include, establish testing authority, and authorize the assessment.
2
Agree the test policyDefine exclusions, testing windows, request limits, activities requiring approval, and the people responsible for those decisions.
3
Name a fix ownerIdentify who receives findings, coordinates remediation, and reports completed fixes for verification.
4
Build the baselineThe ASM and Brand agents establish the initial external view and identify relevant exposures.
5
Start testingThe testing agent assesses exposed systems within the authorized policy, with the first findings reviewed together.
6
Move to testing on changeSubsequent tests follow relevant events and the rolling schedule, while confirmed findings move through remediation and retesting.

The first test can begin once the scope and testing policy are confirmed and authorized. SOC telemetry adds the ability to investigate earlier exploitation, but it is not required to begin external testing.

Read about external penetration testing >

Talk to Our Experts

See how ThreatDefence automates security monitoring, compliance and cyber risk management in one platform.

    I'm an MSP, MSSP, or reseller

    By submitting this form, you accept the terms in our privacy policy.

    Frequently Asked Questions

    What is agentic penetration testing?

    Agentic penetration testing, also called agentic pentesting or AI penetration testing, uses AI agents to plan and execute authorized security tests. Unlike a fixed scanner, the agent can examine a target’s responses and adapt its next testing step. ThreatDefence uses this approach to continuously assess external systems, confirm exploitable weaknesses, and preserve the evidence. SOC analysts review significant findings, while execution remains within the agreed testing policy.

    What does the managed penetration testing service include?

    ThreatDefence’s managed penetration testing service covers authorized internet-facing web applications, APIs, exposed services, identity functions, and cloud resources. The testing agent performs testing and records confirmed weaknesses, while SOC analysts validate significant findings and investigate available telemetry for earlier exploitation. The service also tracks remediation and retests reported fixes. Internal network penetration testing, social engineering, physical testing, and independent assurance reports remain outside the service.

    Which agents take part in a test?

    The ASM agent maintains the external asset inventory, and the Brand agent identifies relevant external exposures, including credential and impersonation findings. The Planner agent determines testing priorities, while the testing agent explores authorized systems, develops test sequences, and proves weaknesses. For environments connected to security monitoring, the Defender and Hunter agents examine available telemetry for signs that a confirmed weakness may already have been exploited.

    Is it safe to run against production systems?

    Yes, within an approved policy designed to limit testing impact. ThreatDefence restricts agents to authorized targets, caps request rates, supports testing windows, and excludes destructive activities such as denial of service or modifying and removing data. Isolated sandboxes enforce network boundaries independently of the AI model. Higher-impact activities require human approval. As with other production security testing, the system owner should review scope and operational constraints before authorizing execution.

    Who is accountable for the testing?

    ThreatDefence’s SOC manages the testing service and reviews significant findings, while your organization authorizes the scope and remains responsible for its systems. The agents operate under controls that define permitted targets and methods, and activities requiring additional authorization wait for human approval. Every test request and result is recorded. Your team handles remediation, and ThreatDefence repeats the relevant tests to verify that reported fixes are effective.

    How does it relate to attack surface management?

    Attack surface management identifies the systems your organization exposes to the internet, including assets absent from existing inventories. Agentic penetration testing uses that information to assess whether weaknesses on authorized systems can be exploited. The ASM agent supplies the changing asset inventory, and the testing agent performs the validation. Newly discovered assets outside the approved scope still require authorization before testing, even when discovery has already identified them.

    How soon is a newly exposed system tested?

    A new or changed system can be queued for testing as soon as the ASM agent records it, ahead of the normal rolling schedule. Priority depends on the exposure and available context, rather than discovery order alone. An active internet-facing application may take precedence over an unused hostname. The service also responds to new vulnerability advisories, reported fixes, and specific testing requests, while remaining subject to approved time windows and testing limits.