LLM application
Prompts, retrieval, output handling, identity, tenancy, data flows, and abuse controls around a chatbot or copilot.
PSC assesses LLM applications, AI agents, and Model Context Protocol integrations as connected systems rather than isolated models. The engagement maps prompts, tools, resources, credentials, memory, and approval boundaries, then tests whether untrusted content or a compromised connector can trigger unintended actions.
This assessment is for teams deploying copilots, autonomous workflows, MCP servers or clients, and tool-enabled agents. The outcome is a prioritized account of exploitable trust failures and practical controls for authorization, consent, isolation, monitoring, and human approval.
The scope follows credible attack and failure paths rather than a generic control checklist.
A clear fit decision protects assessment quality and prevents a broad request from becoming a shallow exercise.
Options can be combined after owners, environments, authority, safety limits, and expected evidence are agreed.
Prompts, retrieval, output handling, identity, tenancy, data flows, and abuse controls around a chatbot or copilot.
Planning loops, memory, delegated tools, credentials, side effects, approval gates, and failure containment.
Servers, clients, transports, tools, resources, prompts, scopes, consent, and malicious-connector scenarios.
Each phase produces an observable input to the final risk and remediation decisions.
Inventory models, prompts, retrieval sources, agents, MCP connectors, tools, credentials, memory, and human decision points.
Define abuse cases for injection, goal hijack, confused-deputy behavior, excessive agency, data disclosure, poisoning, and denial of wallet.
Exercise direct and indirect injection, unsafe output handling, tool chaining, scope bypass, malicious resources, and connector replacement within the rules of engagement.
Test authorization, consent, isolation, allow-lists, approval gates, rate and step limits, logging, and incident reconstruction.
Provide reproducible evidence, system-level remediation, residual-risk decisions, and verification of agreed fixes.
Reporting separates observed evidence, uncertainty, business priority, and the action needed to verify remediation.
Severity combines exploitability, required access, action autonomy, data sensitivity, blast radius, reversibility, and the strength of compensating human or technical controls. Model output alone is not treated as impact unless it crosses a trust boundary or enables a credible harmful action.
PSC presents an executive risk narrative and a technical walkthrough for product, engineering, and security teams. Findings identify the affected trust boundary, required preconditions, evidence, and a testable remediation outcome.
Remediation support focuses on least-privilege tool design, explicit consent, output mediation, isolation, secret handling, observable action trails, and safe failure behavior. Recommendations distinguish application controls from provider-dependent limitations.
An agreed retest verifies corrected authorization, injection resistance, isolation, and approval behavior using the original evidence path. Material architecture changes are re-scoped rather than assumed equivalent.
Timing depends on scope, access, environment, and evidence quality; PSC confirms milestones during scoping rather than publishing unsupported duration promises.
Confirm objectives and authority
Map the connected AI system
Execute controlled abuse cases
Review evidence with builders
Verify prioritized fixes
These publications show the technical reasoning and defensive context behind the engagement.
Trust boundaries, indirect injection, tool use, and agent control design.
Read research →Security implications of connecting an AI assistant to penetration-testing tools.
Read research →Agent attack surfaces across tools, memory, permissions, and orchestration.
Read research →Application-layer risks and defensive controls for generative AI.
Read research →Final rules, access, and deliverables are confirmed in the engagement scope.
Yes. API testing remains relevant, but MCP adds model-mediated tool selection, prompt and resource content, consent, connector trust, and delegated authority. The assessment follows those additional paths from untrusted content to tool effects.
Only with explicit safeguards and rules of engagement. Consequential actions are normally redirected to a controlled environment, test tenant, mock tool, or approval gate so evidence can be collected without creating business impact.
No. The goal is to reduce reachable impact through layered controls, constrained authority, isolation, monitoring, and safe human decisions. Residual model uncertainty is documented rather than hidden.
Provide architecture and data flows, model and agent configuration, tool and MCP inventories, test roles, logging access, and owners who can explain intended workflows and approve safe test cases.
Share the system, decision, and constraints. PSC will confirm whether this service fits before proposing a scope.