Cyber security

Pay for the response,
not the dashboard.

Security vendors sell dashboards, alert counts and certificates. What protects you is how fast a person responds and what that person is allowed to do. Judge any provider on one yardstick, the six functions of the NIST Cybersecurity Framework, and put the same questions to every one. No RFP.

6NIST functions, one yardstick
19Questions to put in writing
1Accountable owner for every decision
0RFPs needed to start

One yardstick, six functions.

The NIST Cybersecurity Framework 2.0 describes security as six functions, worked on together. Govern sits at the center: the policies, roles and risk decisions the rest follow. Providers differ in which functions they do well, so line up each quote against a function before you compare totals.

The six functions of the NIST Cybersecurity Framework drawn as rings: Govern at the center, then Identify, Protect, Detect, Respond and Recover, with who usually delivers each: consulting for Govern and Identify, an MSP or MSSP for Protect, an MSSP for Detect and Respond, an MSP for Recover. (opens the full-size image in a new tab)

Govern

Set the rules, roles and risk appetite.

Who usually delivers it: Security consulting, a virtual security chief.

Ask: Who owns security decisions, and what will you attest to for my insurer?

With AI: Who may use AI, and who answers when it is wrong.

Identify

Know what you have and where it is weak.

Who usually delivers it: Consulting; an MSP for asset lists; vulnerability scanning.

Ask: How do you find assets and weak points, and how often?

With AI: Find every AI tool in use, including the ones staff added on their own.

Protect

Make the doors harder to open.

Who usually delivers it: An MSP (patching, endpoints); an MSSP (firewalls, email, identity).

Ask: What do you manage, what do you only recommend, and who is accountable for each?

With AI: Give each AI agent its own identity and the least access it needs.

Detect

Spot an attack.

Who usually delivers it: An MSSP (round-the-clock monitoring, SOC, EDR, MDR, XDR).

Ask: What do you watch, where do your analysts sit, and can I see the detection rules?

With AI: Ask what the person does and what the AI does when an alert fires.

Respond

Contain and fix it.

Who usually delivers it: An MSSP; incident response retainers.

Ask: What can you do on your own at 3 AM, how fast, and who approves?

With AI: A person approves any AI action that could take a system down.

Recover

Get back to normal.

Who usually delivers it: An MSP (backup and disaster recovery).

Ask: How long to restore, and when did you last test it?

With AI: Keep a way to run the business without the AI.

In general, MSPs cluster at Protect and Recover, MSSPs at Detect and Respond, and consultants at Govern and Identify. Providers differ, so treat this as a starting point.

Who does what: MSP, MSSP and security consulting.

The same request for proposal gets answers from three kinds of provider, and they sell different things. Read across a row. Yes means typically in scope. Varies by provider means ask. Add-on means usually sold separately. Unclear means we could not say. Not typical means do not expect it. This is an abbreviated view; the full guide maps about 40 services.

Who usually delivers each service. MSP means managed service provider; MSSP means managed security service provider.
ServiceNIST functionMSPMSSPSecurity consulting
Virtual security chiefGovernNot typicalNot typicalYes
Compliance help (HIPAA, PCI, SOC 2, CMMC)GovernNot typicalNot typicalYes
Vulnerability management and scanningIdentifyNot typicalYesNot typical
Penetration testingIdentifyNot typicalNot typicalYes
Patching and updatesProtectYesVaries by providerNot typical
Managed firewall and network securityProtectNot typicalYesNot typical
Email securityProtectNot typicalVaries by providerNot typical
Round-the-clock monitoring and alert triageDetectNot typicalYesNot typical
Endpoint detection and response (EDR)DetectNot typicalYesNot typical
Managed detection and response (MDR)DetectVaries by providerYesNot typical
Incident response helpRespondNot typicalYesNot typical
Rapid incident response retainerRespondUnclearAdd-onUnclear
Backup and disaster recoveryRecoverYesNot typicalNot typical
Help deskDay-to-day ITYesNot typicalNot typical

What the six functions leave out: the seams.

The framework and most provider packages are built from boxes: a tool for email, a tool for endpoints, a tool for the firewall. Each box does its job. The attacks that hurt move between the boxes. NIST says the six functions run together, but it does not tell you who owns the places where they meet. Five capabilities live in those seams, and most contracts leave them to you. Read the article on the seams.

Outside the wall

Many security services start watching when something touches your network. Attacks start earlier: probing your website, guessing logins, running bots. Edge services, such as a web application firewall, bot management and denial-of-service protection, can warn you at that stage. Often the web team owns them, and the security provider never sees what they see.

Ask: Do you see what hits my website and logins before it reaches my network?

Data leaving

Most attention goes to what comes in. Data can also leave quietly through ordinary web services and cloud apps. Outbound traffic falls between the network team, the security team and the application team, and nobody may own it.

Ask: Do you watch outbound traffic to web services and cloud apps, and who owns that?

Attacks not in the book

Detection built on known patterns catches the patterns it knows. The costly incident is the one nobody wrote a playbook for. Behavior-based tools exist, and they need tuning and a person who looks when they fire.

Ask: What do you do when the behavior is not in your playbook, and how fast does a person look?

Credentials, for people and machines

Identity connects every function, and it now includes service accounts and AI agents. An AI tool that holds your credentials can use them. In a widely reported 2026 incident, an AI agent that left its test environment used exposed cloud credentials at several services. Accounts differ on what data it touched. Give each agent its own identity, short-lived access, and only what its task needs.

Ask: Which credentials do my AI tools and service accounts hold, who rotates them, and who sees them used?

Who owns the integration

Tools that do not talk to each other leave gaps no one is paid to close. A provider's tools and yours have to work as one system, and someone has to own the joints.

Ask: Who owns the connections between your tools and mine, and what happens when one breaks?

Nineteen questions to put in writing.

Ask every provider the same questions, in writing, before you compare prices. The answers, not the brochure, are the product.

Who responds

  1. Can my team co-manage, and what access do they get?
  2. Can you contain a threat on your own, and under what rules?
  3. Is there a response-time guarantee with penalties? How do you report time to detect and time to respond?
  4. What does your response clock measure: acknowledgement, or containment?
  5. Can you disable a compromised account automatically at 3 AM?

What you can see and keep

  1. Where do your SOC analysts sit?
  2. How long are logs kept hot, how long cold, and who owns them?
  3. Can I see your detection rules and tuning, or is it a black box?

How it fits what you run

  1. What happens to my existing endpoint and SIEM licenses?
  2. Do you cover Mac?
  3. Is pricing split into separate line items?

AI, insurance and proof

  1. What does a person do, and what does the AI do, at each step? Who approves containment?
  2. Can I see your latest SOC 2 report, the period it covers and the systems in scope?
  3. What will you attest to for my insurer, and what must I attest to myself?

The seams

  1. Do you see what hits my website and logins before it reaches my network?
  2. Do you watch outbound traffic to web services and cloud apps, and who owns that?
  3. What do you do when the behavior is not in your playbook, and how fast does a person look?
  4. Which credentials do my AI tools and service accounts hold, who rotates them, and who sees them used?
  5. Who owns the connections between your tools and mine, and what happens when one breaks?

Where AI, insurance and proof fit.

Three things now sit on top of the six functions. Each changes what you ask a provider.

AI cuts both ways

Attackers use AI to write phishing that sounds like your colleagues and to fake voices on payment calls. These are industry forecasts, not measured trends.

Providers use AI to sort alerts faster, and the better ones keep a person in charge of any action that could take a system down. Your own AI tools are a new target too, because a hidden instruction in a document, called prompt injection, can steer them.

So add one test: ask what the person does and what the AI does at each step, and who approves containment. A NIST discussion draft on AI in critical infrastructure, which is not official guidance and may change, reaches a similar view: set limits on what an AI system may do, and enforce them outside the system itself.

See the AI hub

Your insurer will ask

Insurers ask what controls you run, such as multifactor authentication, endpoint detection and backups, and increasingly how you use AI. What you write on the application is a signed statement.

In 2022, an insurer sued to void a policy after a ransomware claim, saying the application claimed multifactor authentication the insured did not have (Travelers v. ICS). Reports differ on how the case ended, so read it as a warning, not a ruling.

Check each answer against what is running, and ask your broker which policy covers a loss that AI causes.

Read the insurance article

Attestation is a promise

An attestation is a formal statement that something is true, made by someone who can be held to it. An independent one comes from a licensed CPA firm: a SOC 2 report is the firm's opinion on a provider's controls, and it is a report, not a certificate.

A Type 1 report describes the controls on one date. A Type 2 report tests them over a period. A self-attestation is one you sign yourself, on an insurance application or, for defense contractors, as the annual affirmation under CMMC.

Ask each provider for its latest SOC 2 report and what it covers. Treat anything you sign as a promise.

Provider types, not names.

Providers describe themselves in a few ways. Treat each description as their claim, and test it.

Global-scale providers

Position themselves for large, regulated organizations.

Test: will they fit a smaller environment, and at what price?

Cloud-native providers

Position themselves for fast deployment and heavy automation.

Test: false positives, and how they handle older systems.

Legacy and compliance specialists

Position themselves for mainframes, industrial systems and audit support.

Test: how much is automated, and how fast they respond.

Service-led mid-market providers

Position themselves on named teams and threat hunting.

Test: global coverage and scale.

Small regional providers

Position themselves for small, lower-risk environments.

Test: what happens as you grow.

Consulting add-ons

Assess and advise, but do not watch.

Test: who watches at night.

Reading a price sheet. Providers quote per endpoint, per device, per platform or per IP address. Convert every quote to one unit before you compare. Then ask whether the quote splits into separate line items, and what is left out of the service.

Definitions in plain words.

Every term on this page, in one place.

Attestation
A formal statement that something is true, made by someone who can be held to it. An independent one comes from an outside firm. A self-attestation is one you sign yourself.
CIS Controls
A prioritized list of security safeguards published by the Center for Internet Security. Providers use it for practical, step-by-step detail.
CMMC (Cybersecurity Maturity Model Certification)
A US Department of Defense program. Defense contractors must meet cybersecurity requirements and show it, by self-assessment or by an outside assessor, depending on the level.
Co-management
You and the provider both have access to the same security tools and share the work.
Containment
Stopping an attack from spreading, for example by isolating a computer or blocking an account.
Dwell time
How long an attacker is inside your systems before anyone finds them.
EDR (endpoint detection and response)
Software on computers and servers that records activity and flags or acts on threats. Someone still has to watch it.
Endpoint
A computer, server, phone or other device that connects to your network.
Eradication
Removing the attacker and whatever they left behind.
Forensics
Investigating what happened, and preserving the evidence.
Hot and cold log storage
Logs are the records your systems keep. Hot storage keeps recent logs ready to search fast. Cold storage keeps older logs more cheaply and retrieves them more slowly.
Incident response
The plan and the work for handling an attack, from first alert to recovery.
ISO 27001
An international standard for running an information security management program. Organizations can be audited and certified against it.
MDR (managed detection and response)
A service where another team watches your systems around the clock, investigates alerts and acts for you.
MFA (multifactor authentication)
Logging in with something more than a password, such as a code from your phone.
MITRE ATT&CK
A public catalog of the tactics and techniques attackers use, used to test whether detection covers real attacker behavior.
MSP (managed service provider)
A company that runs your day-to-day IT for a recurring fee: help desk, patching, backups.
MSSP (managed security service provider)
A company that runs security for you, usually through a round-the-clock security operations center.
MTTD and MTTR (mean time to detect and respond)
Averages of how long it takes to spot a threat and to respond to it. Ask how each is measured, because "response" can mean acknowledged or contained.
NIST AI Risk Management Framework
Voluntary US guidance for managing AI risk, organized as govern, map, measure and manage.
NIST Cybersecurity Framework 2.0
A voluntary US government framework that describes security as six functions: govern, identify, protect, detect, respond and recover. It is a common vocabulary, not a certification.
NIST profile
A custom list of the security outcomes an organization wants from each function, used to measure the gap between where it is and where it wants to be.
NOC (network operations center)
The team that monitors and maintains your network and systems for availability, as opposed to security.
Penetration test versus vulnerability scan
A scan is automated and checks for known weaknesses. A penetration test is people carrying out an authorized simulated attack to prove what can be broken.
Phishing and business email compromise
Phishing is a fake message meant to trick you. Business email compromise is a fake message, often posing as an executive or a vendor, meant to get money or data.
Prompt injection
Hidden instructions in text, a document or a web page that steer an AI system into doing something its owner did not intend.
RACI
A table that names who is Responsible, Accountable, Consulted and Informed for each decision.
Remediation
Fixing the root cause and the weakness, not only stopping the attack.
RIR (rapid incident response)
Expert help during an active attack: containment, forensics and recovery, often bought as a retainer.
RMM (remote monitoring and management)
Software an MSP uses to watch, patch and manage your devices remotely.
Service level agreement (SLA)
The written promise about what the provider will deliver, how fast, and the remedy if it does not.
SIEM (security information and event management)
A system that collects logs from many sources, connects the dots and raises alerts.
SOAR (security orchestration, automation and response)
Software that runs the repeatable steps of a response automatically, using prewritten playbooks.
SOC (security operations center)
The team, and sometimes the room, that monitors for threats around the clock.
SOC 2
A report by an independent CPA firm on a service provider's controls, tested against standard criteria. A Type 1 report describes the controls at a point in time. A Type 2 report tests them over a period.
Threat hunting
People searching for attackers that automated tools missed.
Threat intelligence
Information about attackers, their methods and what they are targeting now, used to improve detection.
vCISO (virtual chief information security officer)
A part-time or outsourced security leader who sets strategy, policy and priorities.
XDR (extended detection and response)
A platform that watches endpoints, network, cloud, email and identity together, so it can connect clues across them.
Zero trust
A design principle: no user, device or system is trusted by default, and every request is checked.

Why you can trust what you read.

Suppliers don’t pay to be listed. If you buy through us, the supplier you choose pays IQ4hire, and we tell you before work begins. Need our experience on its own, such as an assessment or a provider review? That is consulting, on a fee. Either way you see every realistic option, including keeping what you have.

We have run this, not just read about it. IQ4hire has live security engagements and recently completed a full evaluation across managed service, managed security and consulting providers. Our infrastructure lead has owned network and perimeter security since 1986.