The Agentic Era Security Program
The approach behind every Armada engagement: start from business context, test continuously, work from one action plan and prove it holds with red team validation twice a year.
Fight fire with fire.
For two decades, security testing meant an annual penetration test, a quarterly scan and a report that went stale the day it arrived. That model assumed an attacker who moved at human speed. AI has to become the engine of the testing program itself: agents for coverage and repetition, people for judgment.
Eight rules we don't break.
- 01Obsess over fundamentals.Least privilege, secure configuration, patching and a sound SDLC still decide most outcomes.
- 02Coverage before cleverness.You can't defend what you don't know exists. Inventory comes before advanced testing.
- 03Continuous, not point in time.Attackers using AI run every day, so testing must too.
- 04Everything is an exposure.A host, an MCP server, an attack path and a compromised account are all managed the same way.
- 05Business context sets priority.Domain admin always matters. Beyond that, what matters is what makes your business run.
- 06Fight fire with fire.Autonomous agents for coverage. Human red teams for creativity, stealth and response.
- 07Humans own the anomaly.Automation handles the baseline. People are paged when something departs from it.
- 08Every finding ends in an action.Data that doesn't change what someone does tomorrow is noise.
Nine steps to a continuous exposure program.
We always start in the same place: what your business can't afford to lose, and what you actually have. Everything else builds on those first two steps.
- 01Foundation
Establish business context
Name the 5 to 10 critical assets the business can't afford to lose, record where each lives and what feeds it, and get written executive sign-off so priorities hold when work is delegated.
- 02Foundation
Build the map
Inventory infrastructure, applications and APIs, cloud, SaaS, identities and AI assets, with network and data flow diagrams. An unowned asset is a finding.
- 03External exposure
Run continuous external exposure management
Discover everything an attacker can see from the internet, alert on new assets within hours, and validate exploitability instead of forwarding scanner output.
- 04Internal testing
Deploy continuous internal agentic testing
Place AI testing agents in every site, segment and trust zone, under rules of engagement you approve, and test AI assets as their own class.
- 05Prioritize
Prioritize attack paths
Merge external and internal findings into one attack-path graph. Paths to administrative control first, then to critical assets, then through AI assets. Find the choke points.
- 06Active defense
Deploy active defense
Adaptive deception agents that mislead and expose attacker agents, placed in every segment that hosts a testing agent, and an inline AI Gateway that stops prompt injection and keeps your own agents on approved models, MCP servers and tools.
- 07AI usage
Monitor behavioral exposure through AI usage telemetry
Baseline each user's AI activity and alert on departures, so compromised accounts and insider risk surface early.
- 08Converge
Unify findings into one action plan
Normalize, de-duplicate and rank every finding. Each action has an owner, an effort estimate and an expected risk reduction, and is retested to closure.
- 09Validate
Validate with twice-yearly red team engagements
Human-led red team and AI red team engagements against agreed objectives test whether people, detections and response stop a skilled adversary. Gaps become new automated tests.
The first 90 days, with a gate at every phase.
Larger or more regulated enterprises should expect each phase to run longer. The first red team engagement follows shortly after day 90, so it tests the live program.
- DAYS 1–30 · STEPS 1–2
Foundation
Critical assets, asset inventory including AI assets, network and data flow diagrams.
GateCritical assets signed off. Every known asset has an owner.
- DAYS 31–45 · STEP 3
Outside-in
Continuous external exposure management.
GateNew external assets alerted within hours and reconciled to the inventory.
- DAYS 46–75 · STEPS 4–6
Inside-out
Internal testing agents, deception and the AI Gateway in priority segments. Attack-path ranking.
GateRules of engagement approved. Agents and decoys live in every priority segment. Agents that touch critical assets route through the AI Gateway.
- DAYS 76–90 · STEPS 7–8
Converge
Behavioral telemetry baselines and the unified action plan.
GateBaselines built. Every top-priority action has an owner and a retest date.
Measure shrinking exposure, not how much you found.
- Inventory coverage: the share of discovered assets you already knew about
- AI assets discovered versus AI assets inventoried
- Time to discover a new internet-facing asset
- Open paths to administrative control and to each critical asset
- Mean time to remediate, by action-plan priority, and retest pass rate
- Deception engagements and time to human alert
- Prompt-injection attempts stopped at the AI Gateway, and agents routed through it
- Red team objectives reached, and findings continuous testing missed
Built on standards your auditors know.
- Gartner CTEM for the scope, discover, prioritize, validate and mobilize cycle
- MITRE ATT&CK for internal testing coverage
- MITRE ATLAS for adversary techniques against AI assets
- OWASP Top 10 for Agentic Applications 2026 for agents that plan, use tools and hold memory
- OWASP Top 10 for LLM Applications for model-level risks such as prompt injection
Take the full playbook with you.
Every step with a practitioner checklist, the 90-day roadmap, the metrics and the framework mappings, in one PDF you can hand to your team.
The Agentic Era Security Program Playbook
How to run continuous exposure management at the speed attackers now move
Armada2026
Get the playbook
We'll email you the PDF. No sales call unless you ask for one.
Check your inbox.
The playbook is on its way. If you'd like to walk through it with a member of our team, get a demo.
Start with what you can't afford to lose.
We'll work through your critical assets and your asset map with you, then show you what the rest of the program would add.