Defend

Turn red-team findings into controls engineering teams can ship and security teams can verify.

Deterministic controls, release gates, telemetry, and retest evidence. Defend hardens guardrails, evals, approval workflows, permissions, and release gates — and produces the retest criteria that close the red-to-blue loop.

Harden Agentic Workflows View 8-Service Portfolio

Boardroom to backlog evidence chain diagram

Controls to backlog

Defend visual operating model

Buyer questions

- What controls need to exist before launch?
- Are guardrails, evals, release gates, telemetry, and approval workflows meaningful enough?
- Can approval boundaries be bypassed?
- Do logs and telemetry support investigation and evidence?
- What must become Jira or Confluence work?

Products / instruments

SecEng Model Gateway — Governed AI Routing & Spend Control SecEng Runtime Proxy — Trace, Replay, Evidence Export SecEng Authority Graph — Agent Blast-Radius Map SecEng Surface Scanner SecEng RAG Test Harness — XPIA Lab SecEng Program Blueprint Kit

Featured research

AI Security Engineering Handbook AI Product Security in the Age of Mythos

Featured research

AI Security Engineering Handbook

A structured study companion for AI security engineering practitioners, candidates, and teams covering foundations, role architecture, domain vocabulary, controls, evidence, assessment, and operating-model design.

14 ChaptersStudy Companion14 AIPSA domains

Open report

Defend instrument

SecEng Runtime Proxy

Delegated authority

SecEng Authority Graph

Evidence proof

AI Governance Controls: Garak, Nemo, Presidio, Promptfoo

Guardrails, evals, control mapping, and implementation backlog evidence.

View proof

Flagship

DefendAvailable

implementation

Agentic Workflow Security & Hardening

A hardening engagement for AI agents and workflows: permission design, tool policies, approval gates, scoped credentials, logging, rollback, exception handling, and production guardrails.

Outcome

5 deliverables

Best for

AI Platform Lead, Product Security, Security Architect, Automation Lead

•Tool permissions, scoped credentials
•Human approval, destructive actions
•Logging, kill switches, rollback
•Incident hooks, CI/CD gating checklist

Duration: 3-6 weeksScoped through no-cost scoping

View service See people fit View evidence

Flagship

DefendAvailable

assessment

AI Guardrails & Evals Review

A review and improvement plan for guardrails, evals, refusal behavior, fallbacks, test coverage, monitoring, regression cases, and release criteria.

Outcome

5 deliverables

Best for

AI Product Lead, Product Security, Trust and Safety, Engineering Lead

•Guardrail architecture, safety policy, refusal, fallback, and monitoring review
•Eval suite, abuse case, failure mode, and regression coverage review
•Prompt/control regression testing and release quality gate recommendations
•Engineering-ready remediation plan for guardrails, evals, and release criteria

Duration: 2-5 weeksScoped through no-cost scoping

View service See people fit View evidence

Sample deliverables

Control acceptance criteria

Red-team retest criteria

Release gate recommendations

Guardrail / eval hardening guidance

Telemetry and alerting requirements

Rollback and approval paths for agentic workflows