For MSPs, MSSPs and VARs

Can your team deliver AI security?

Eighteen questions, about eight minutes. See where your practice can already say yes to an AI security engagement, and where you would need more bench to deliver it.

How to use it

Answer for your delivery team as it is today, not the hires you are planning.
Count only skills your team has used on a real engagement, not certifications alone.
The weakest dimensions are where a client request would stall.

Your answers stay in this browser. Nothing is sent or stored unless you choose to contact us.

Offer & Demand0/3
Workforce AI & Data Protection0/3
AI Governance & Risk0/3
Identity, PAM & Agent Access0/3
AI Application & Runtime Security0/3
Operations & Capacity0/3
Your results →

Offer & Demand

Dimension 1 / 6

Offer & Demand

Whether AI security is something you sell, or something clients ask about and you improvise.

1. Do clients ask you about AI security or AI governance?

0Not yet. 1Occasionally, without a structured answer. 2Regularly; we answer case by case. 3Regularly, and the questions feed a defined offer.

What it means: Client questions are demand. Capture them in discovery and route them to one defined offer instead of answering case by case.

2. Do you have a packaged AI security offer?

0No. 1Ad hoc consulting only. 2A defined assessment, without follow-on services. 3An assessment that leads into a priced roadmap of follow-on work.

What it means: Lead with an assessment that ends in a roadmap: it is easier to sell than a point product and it opens the follow-on projects.

3. Can your sales team scope an AI security engagement?

0No. 1Only with an engineer in every call. 2For simple cases. 3Yes, with discovery questions, scoping templates and estimates.

What it means: Give sales a short discovery script and scoping template so an engineer is needed for the design, not for every first conversation.

Workforce AI & Data Protection

Dimension 2 / 6

Workforce AI & Data Protection

Discovering and controlling the AI tools a client's employees already use.

1. Can you discover the AI apps and extensions in use at a client?

0No. 1From a survey or interviews. 2With a tool, for sanctioned apps. 3With a tool, including unsanctioned apps and browser extensions.

What it means: Discovery is the first deliverable of any workforce AI engagement; without a tool to do it, the rest of the policy is guesswork.

2. Have you deployed controls on prompts and uploads (AI data loss prevention)?

0Never. 1In a lab. 2At one client. 3At several clients, including policy tuning after go-live.

What it means: Prompt and upload controls need tuning after go-live; a first deployment is where most of the learning, and the risk to the client relationship, sits.

3. Can you write an AI acceptable-use policy with a client?

0No. 1From a generic template. 2Tailored to the client's data and tools. 3Tailored, enforced by technical controls and reviewed on a schedule.

What it means: A policy is only as useful as the controls behind it; pair the document with discovery and data-protection controls.

AI Governance & Risk

Dimension 3 / 6

AI Governance & Risk

Helping a client decide who is accountable for AI and how risk is assessed.

1. Can your team run an AI risk assessment against a recognised framework?

0No. 1Informally. 2Against one framework, on paper. 3Against a framework such as the NIST AI RMF, ending in owners and controls.

What it means: Clients increasingly expect a recognised framework behind an AI risk assessment; NIST AI RMF is the common reference in the US.

2. Can you set up AI governance: owners, approvals and change control?

0No. 1We would advise in general terms. 2We can design it. 3We design it and connect it to the client's change management.

What it means: Governance that is not connected to change management stops working the first time someone adds a tool or a data source.

3. Do you have fractional CISO or senior advisory capacity for AI?

0No. 1Occasionally, from one person. 2Yes, with limited availability. 3Yes, with capacity to run a 12–18 month program.

What it means: AI programs need senior judgement over months; one stretched advisor becomes the bottleneck of the whole practice.

Identity, PAM & Agent Access

Dimension 4 / 6

Identity, PAM & Agent Access

The identity work every AI agent depends on, from IGA and PAM to agent identities.

1. How deep is your IGA and PAM delivery experience?

0None. 1Basic MFA and SSO. 2One IGA or PAM platform. 3Two or more of CyberArk, Delinea and SailPoint, in production.

What it means: Agent and non-human identity work is built on mature IGA and PAM; without that depth, NHI projects inherit the gaps.

2. Can you design identities and least-privilege access for AI agents?

0No. 1In principle. 2For simple integrations. 3Including delegation, scoped tokens, MCP authorization and tested revocation.

What it means: Agent access design — delegation, scoped and short-lived credentials, tested revocation — is new for most teams and quick to get wrong.

3. Can you staff a 12–18 month identity program?

0No. 1Part-time, with one engineer. 2With a small team, at risk if someone leaves. 3Yes, with a team that covers the whole program.

What it means: Identity programs run for a year or more; a single engineer is a continuity risk the client will notice.

AI Application & Runtime Security

Dimension 5 / 6

AI Application & Runtime Security

Protecting AI applications and agents while they run: injection, tools, MCP servers.

1. Can you assess an AI tool or MCP server before a client adopts it?

0No. 1We read its documentation. 2We run a scanner. 3Provenance, code and isolated behaviour testing, with an approval record.

What it means: Admission reviews need more than a scan: provenance, code where available and behaviour in an isolated environment.

2. Have you deployed AI runtime protection (prompt-injection and output inspection)?

0Never. 1In a lab. 2At one client. 3At several clients, with coverage mapped to every data path.

What it means: Runtime protection only covers the paths it sits on; mapping those paths is half of the engagement.

3. Can you test an AI workflow against misuse (red-team style)?

0No. 1Ad hoc. 2With a test set for one workflow. 3With repeatable misuse and legitimate-use test sets, rerun after changes.

What it means: A repeatable test set is what turns a one-off assessment into an ongoing service.

Operations & Capacity

Dimension 6 / 6

Operations & Capacity

Running AI security after go-live, and having the people to do it.

1. Can your SOC or NOC monitor and respond to AI agent incidents?

0No. 1Only through existing alerts. 2With some AI-specific detections. 3With AI-specific detections, containment runbooks and exercises.

What it means: AI incidents need their own runbooks: disabling an identity is one step among several, and each needs an owner.

2. How many engineers could you put on a new AI security engagement next month?

0None. 1One, part-time. 2One or two, by delaying other work. 3A dedicated team, without delaying other work.

What it means: Capacity is what turns a won deal into a delivered one; a sale that waits for a hire is a sale at risk.

3. Do your engineers hold current vendor certifications for the platforms involved?

0None. 1A few, on some platforms. 2On our main platforms. 3Across the platforms we sell, kept current.

What it means: Vendor certifications matter in partner programs and to clients; plan them with the platforms you intend to sell.

Your results

Scores run from 0 to 100 per dimension. Low scores are not a verdict: they are where to add capacity before you sell.

Answer every question to see the full results. Unanswered questions count as 0.

Where to add capacity first

Fill the gaps without new headcount

Telekomplex extends your bench with certified engineers for each of these practices, delivered white-label inside your delivery model. Bring us the dimension that scored lowest.

Talk to Telekomplex