AI pentesting, or AI-powered penetration testing, uses AI to explore attack paths and test whether unauthorized access or actions are possible. Services differ in the steps they automate and the work that security engineers perform.
With Layer8 Trident, our in-house AI explores and tests web applications and APIs. Security engineers verify reproducibility and impact before reporting findings. We agree on scope and operating conditions before testing begins and deliver confirmed findings with evidence.
Testing with AI and testing an AI system
The terms “AI vulnerability assessment” and “AI penetration testing” can describe different services:
- Testing performed with AI: AI analyzes source code or interacts with running applications and APIs to investigate vulnerabilities.
- Testing an AI system: An assessment examines risks such as prompt injection in systems that use LLMs or AI agents.
Trident uses AI to test web applications and APIs for exploitable attack paths and their impact. If your objective is to test risks specific to an LLM or AI agent, confirm that scope with the provider separately.
A broad assessment for vulnerabilities and an engagement designed to test a specific data-theft or unauthorized-action scenario can require different scope and methods. Tell the provider which risks matter and what evidence you need.
Questions to ask a provider
| Topic | What to confirm |
|---|---|
| Scope | Does testing cover web apps, APIs, networks or AI systems? Can it exercise authenticated features and different permission levels? |
| Inputs | Which URLs, test accounts, API specifications or source code are required? |
| AI and human responsibilities | Who defines scope, runs testing, confirms reproducibility and impact, and approves the report? |
| Evidence and limitations | Are reproduction steps, traffic, impact and remediation guidance included? Are untested areas and safety-related exclusions documented? |
| Operating controls | How are scope, traffic limits, prohibited actions, testing windows and stop procedures enforced? |
| Retesting | What findings, number of checks and deadlines are included? When does additional testing incur a fee? |
| Data handling | What reaches external AI services? What are the training-use, retention and deletion terms? |
Providers combine scanners, AI agents and manual testing in different ways. Evaluate whether the service tests the scope you need and provides findings your team can act on, as well as how many findings it produces.
Evidence from real-world testing
Layer8’s research account, using Trident, ranked first worldwide in HackerOne’s VDP Individual category for Q3 2026. From July to September 2026, 568 reports were triaged across 50 programs.
Our technical write-up describes an attack path that combined an authentication bypass with SQL injection. It also explains the roles of automated exploration, attack validation and human report review.
These are results from a research account. They do not guarantee a particular number of findings in a customer engagement or the discovery of every vulnerability. Comparisons need matching targets, evaluation periods, duplicate and false-positive handling, and acceptance criteria.
We began this research in May 2025. Our first field-test report records the initial benchmark results and real-world testing conditions.
Scope, pricing and preparation
Standard Trident assessments are quoted individually. Pricing and timing depend on application and API size, authentication, scope and operating conditions. One-off and monthly engagements include an assessment report and one remediation check of the reported findings.
For an initial discussion, share what you know about:
- Target URLs, the relationship between the application and its API, and whether staging is available.
- Risks to assess, test-account availability and permission levels.
- Preferred timing, release dates, report recipients and required formats.
- Prohibited actions, testing windows and data-handling constraints.
Do not put credentials in the contact form. We arrange a secure handover after agreeing on the engagement. See Trident’s service terms and the sample report overview for details.
Choose a Layer8 engagement
- A one-off or monthly web and API assessment: Layer8 Trident, with scope and operating conditions agreed individually.
- Testing your company’s internet-facing assets: Trident Live-Fire Test, a fixed-price pilot that runs for approximately one month after confirming the targets. The service page lists pricing, availability and exclusions.
- A defined assessment for SOC 2 audit evidence: SOC 2 Application Pentest, covering one web application and its first-party API, with English reports and one retest. The engagement runs by email. It does not guarantee SOC 2 compliance or acceptance by your auditor.
You can contact Layer8 before your scope or timing is final.