PhishRoot PenTest

Saved requests go to your organization's Operations queue for coordination. Execution requires a connected provider, written rules of engagement and separate approval.

Loading organization…

02 / Service scope

Scoped. Authorized. Tested by people.

• Scope & authorization

  Available

  Record internal or external targets, exclusions, an authorized contact and the preferred testing window.

• Engagement tracking

  Available

  Save a scope request to your selected organization's Operations queue and follow its status.

• External posture assessment

  Available

  Use the separate ownership-verified, non-intrusive assessment for public DNS and web signals.

• Network attack validation

  By quote

  Internal and external attack simulation, segmentation checks and evidence-backed attack paths.

• Recurring network tests

  By quote

  Tester-led monthly or quarterly engagements after an approved scope and agreed rules of engagement.

• Executive & technical reports

  Planned

  Provider evidence, remediation recommendations and retest results synchronized to the client workspace.

Testing with clear boundaries

A vulnerability or posture assessment identifies signals; a penetration test validates compromise paths through authorized attack simulation. Our existing assessment remains non-intrusive and is not an exploit-based penetration test. Active testing will require approved targets, exclusions, stop conditions and a separately authorized provider engagement.

No compliance certification, response-time guarantee or live internal sensor is implied. Pricing lists proposed package starting prices, not available testing services. Final scope, currency and commercial terms are confirmed through your MSP or MSSP.

03 / AI & LLM security · By quote

AI & LLM security testing

Security for the application around the model: chatbots, copilots, retrieval-augmented generation (RAG) and tool-using agents. The proposed scope examines how information, instructions and permissions move through the whole AI stack—not just the model in isolation.

AI/LLM penetration testing is delivered by PhishRoot testers, scheduled after scope approval and written authorization. Discuss requirements directly or through your MSP or MSSP; an enquiry does not start an AI test. The existing AI-assisted passive domain assessment does not test LLM applications.

Coverage across the AI stack

• Models & guardrails

  Model configuration, safety boundaries and the controls that separate trusted instructions from untrusted inputs.

• Prompts & context

  User messages, uploaded files, retrieved content and tool responses that can influence the application's instructions.

• RAG & retrieval

  Document access, source integrity, vector-store permissions and separation between customer knowledge bases.

• Agents & tools

  Tool permissions, approval gates and limits on what an agent can read, change or trigger.

• Outputs & connected applications

  How model responses are validated before they reach a browser, an API or another system.

• Infrastructure & APIs

  Authentication, tenant boundaries, key handling, sensitive logging and usage controls around AI services.

04 / AI risk categories

OWASP Top 10 for LLM Applications

OWASP Top 10 for LLM Applications supplies the risk categories below. MITRE ATLAS can inform AI threat scenarios and NIST AI RMF can inform governance discussions. These are planning references only: framework-mapped reports are not implemented, and PhishRoot makes no accreditation, certification or auditor-acceptance claim.

• LLM01

  Prompt injection

  Untrusted messages or documents redirecting application behaviour.

• LLM02

  Sensitive information disclosure

  Private information appearing in responses or crossing customer boundaries.

• LLM03

  Supply chain

  Risks introduced by models, dependencies and third-party components.

• LLM04

  Data & model poisoning

  Untrusted training or retrieval sources changing system behaviour.

• LLM05

  Improper output handling

  Model responses reaching downstream systems without validation.

• LLM06

  Excessive agency

  Agents having more permissions or autonomy than their task requires.

• LLM07

  System prompt leakage

  Exposure of hidden instructions or sensitive configuration.

• LLM08

  Vector & embedding weaknesses

  Retrieval permissions and isolation failures in shared knowledge stores.

• LLM09

  Misinformation

  Unsupported or misleading output driving consequential decisions.

• LLM10

  Unbounded consumption

  Missing usage limits allowing runaway requests or unexpected costs; no load or denial-of-service testing is offered.

05 / How an engagement runs

Scope first. Evidence before conclusions.

1. 01

   Define scope & trust boundaries

   Discuss the AI application, data sources, customer boundaries and permitted actions with your partner.

2. 02

   Agree safe engagement conditions

   Any future testing requires a qualified provider, separate written authorization, exclusions and stop conditions. Prefer isolated staging, synthetic data and restricted tools.

3. 03

   Validate evidence

   PhishRoot testers distinguish reproducible security issues from variable model behaviour, only within the approved scope.

4. 04

   Report, remediate & review

   Proposed reports would connect business impact to technical evidence and recommended fixes. Retest scope and timing must be agreed separately.

06 / Deliverables

Clear findings for engineers and leadership

• Technical findings

  Proposed evidence-backed findings with affected components, severity rationale, relevant risk categories and remediation guidance.

• Executive summary

  Proposed business-risk overview covering impact, priorities and remaining risk.

• Controlled validation evidence

  Proposed sanitized reproduction records from an authorized provider, without customer secrets or unsafe actions against production systems.

• Remediation review

  A future, separately agreed provider review of fixes—not a free retest, certification or guaranteed outcome.

07 / Common questions

Before you plan an engagement