Fee
Register to see indicative feesConfirmed in writing after scoping01 / Preparedness · Supporting
AI Incident Response & Monitoring
If an AI system in your organisation behaved badly tomorrow, you would not have a plan for it, and the security incident plan you do have would not fit.
Defend & RespondPreparedness
Typical duration
5 to 8 weeksScoped in a discovery workshopPrimary artefact
AI incident playbooksPlus 1 moreWho leads it
Steve SharmaFounder & Principal04 / What this is
What AI Incident Response & Monitoring is
Preparation for AI-specific incidents: a model producing harmful output at scale, an agent taking an unintended action, data leaking through a model, or a vendor model changing behaviour without notice.
And what this service produces
It produces playbooks your existing incident function can run, and a detection design that surfaces these events at all. Most organisations would currently not know.
05 / Who it is for
Built for a real assurance trigger, not a badge project.
01
This fits when
- Organisations with AI in customer-facing or operational use.
- Incident response functions with no AI-specific playbooks.
- Anyone whose regulator or insurer expects a documented response capability.
- Teams that have already had a near-miss.
02
The usual trigger
- An AI system did something unexpected and the response was improvised.
- A regulator or insurer asked about AI incident capability.
- Customer-facing AI has gone live.
- An existing IR plan was reviewed and AI was absent from it.
06 / The obligation
What an AI incident looks like, and why it costs when it arrives
The events this prepares for are not hypothetical, and they share a shape: they are visible to customers before they are visible to you.
Harmful output at scale
A model producing harmful or false output at scale.
Irreversible agent action
An agent taking an irreversible action across many records.
Cross-customer data exposure
Customer data surfacing in another customer's session.
Silent vendor model change
A vendor silently updating a model so that a system which passed testing no longer behaves the way it was approved to.
The conventional process does not fit
What makes these expensive is not the technical failure. There is no patch, the blast radius is often unclear for days, containment may mean turning off a revenue-generating feature, and the disclosure question is genuinely hard. Organisations that have rehearsed these decisions make them in hours. Organisations that have not make them in weeks, publicly.
The cost of this preparation is small and knowable. The cost of improvising it during a live incident is neither.
07 / What you receive
Two artefacts. Each one yours to keep.
- 01
AI incident playbooks
Scenario-based playbooks your existing IR function can run, with decision rights and thresholds.
Playbook setIncident manager - 02
Detection design
What to log, what indicates an AI incident, and where the alert goes.
Detection designOperations and engineering
08 / Scope
The whole system is visible before work begins.
Scenarios
- AI-specific incident scenarios defined
- Blast radius and containment options per scenario
- Decision rights and thresholds
Playbooks
- Response steps integrated with existing IR
- Containment options including the commercial trade-offs
- Communications and disclosure considerations
Detection
- Logging requirements
- Indicators of AI incident
- Alert routing
Exercise
- One tabletop exercise against the playbooks
- Findings and revisions
09 / Included / not included
A useful boundary is part of the deliverable.
Included
- Scenario definition
- Playbook design
- Detection design
- One tabletop exercise
Not included
- Monitoring operation: We design detection. We are not a SOC and we do not run it.
- Incident response retainer: We do not hold a pager. If you need someone on call, we will point you to a provider who does.
- Tooling implementation: We state logging and detection requirements. Building them is separate.
10 / How we do it
A traceable path from scope to evidence.
Scenario definition
The incidents that would actually matter for your systems, and the decisions each forces.
Playbook design
Response steps integrated with your existing IR process, not parallel to it.
Detection design and exercise
Logging and indicators designed, then one tabletop against the playbooks.
Mapped againstNIST AI Risk Management FrameworkNIST SP 800-61ISO/IEC 42001
11 / Timeline
5 to 8 weeks of work.
Elapsed time and customer effort are separated because they are not the same number.
Ready
What exists after this engagement
Scenario-based playbooks integrated with your existing IR process, a detection design, and one exercised run-through. The engagement itself runs 5 to 8 weeks.
Customer effort
A scenario workshop, playbook reviews and one tabletop exercise.
Trigger
What activates them
Any AI system behaving outside its approved parameters, a suspected data exposure through a model, or an agent action outside its permission envelope.
Customer effort
Your incident function invokes the playbooks.
Cadence
What keeps them current
We recommend a tabletop every six months and a playbook review on any material model or vendor change.
Customer effort
Running those is yours, or a separate arrangement.
Not us
What happens on the day
Your incident function runs the response.
Customer effort
We are not on call and we do not hold a pager. If you need that, we will point you to a provider who does.
Our work: 5 to 8 weeks
12 / Scope check
Test whether AI Incident Response & Monitoring
is the right next move.
We will cover likely scope, cost and fit.
Book a confidential strategy sessionWith Steve Sharma, who would lead.13 / Who leads the engagement
Named lead. Direct assessment.
Your incident readiness lead
Steve Sharma
Founder & Principal
Facilitates the scenario workshop and runs the tabletop exercise.
ASD-endorsed IRAP AssessorCISSPCREST
Every engagement has a named lead. Where a second executive or specialist is needed, they are named in the proposal and subject to client approval.
Review leadership record14 / Why Cybersecurity Link
Three decisions.
Accountability.
Playbooks that attach to your existing IR function.
A parallel AI incident process is one nobody runs. Ours integrates with what already works.
The exercise is included, not optional.
An unrehearsed playbook is a document. One tabletop is in the scope.
We say plainly that we are not on call.
Most providers leave that ambiguous. We would rather name it and point you to someone who does hold a pager.
15 / The boundary
What this engagement does not claim.
On the day
We prepare. We are not on call.
We do not hold a pager, and we do not run your response. Your incident function does.
Operations
We are not a managed security service.
We design detection. We do not operate monitoring.
Tooling
We design detection requirements.
Implementing the tooling is separate work.
What moves the fee
With the number of scenarios in scope, confirmed at discovery.
Substantially, where a functioning incident response process already exists.
Discovery is a bounded fixed feeEngagement quoted in writing after discoveryOne tabletop exercise included
18 / How this fits
Where this sits in the wider programme.
AI Security Architecture & Threat Modelling
Detection design is far easier where the architecture already supports it.
AI Agent Security
If the concern is agent actions specifically, preventing is cheaper than responding.
Incident Response & Resilience
The broader cyber incident capability this sits inside.
19 / When this is not right
Sometimes the best recommendation is a smaller service, or a later date.
20 / FAQ
The questions that change the decision.
No. We prepare; your incident function responds. We do not hold a pager and we would rather say so here than leave it ambiguous. If you need an on-call provider we will point you to one and take nothing for the referral.
Not separate: integrated. A parallel process is one nobody runs under pressure. The playbooks slot into your existing IR structure with AI-specific scenarios and decision points.
Then this is the wrong starting point and we would say so. The broader resilience service builds the function; this adds AI scenarios to one that exists.
Yes, one. An unrehearsed playbook is a document rather than a capability, so we would rather include it than offer it as an upsell.
21 / Next move
If AI Incident Response & Monitoring is what you need, the next step is a conversation about scope.
A 45-minute conversation with Steve Sharma, who would lead the engagement: what your scope actually looks like, what it would cost, and whether this is the right service for you at all.