01
Govern
If AI has been deployed faster than it’s been governed, and someone has just been made accountable for the gap, this is the group.
Usually starts with AI Governance Starter: the two-week minimum viable version.AI Security
AI Security
Your organisation has deployed AI faster than it has governed it, and someone has now been made accountable for the difference.
This is the practice that closes that gap: nine services covering how AI is governed, how exposed you currently are, and how the systems themselves are defended.
Every service, with pricingEvery AI Security service, with what it costs
Indicative fees are AUD, excluding GST, shown to registered users and inherited from the pricing page. From means the figure moves with scope; Discovery means it’s scoped in a workshop before anything is quoted.
ServiceWhat it producesDurationModeFee
AI Risk Self-Assessment
A maturity indication across five AI governance domains5 minFreeFreeAI Governance & Responsible AI
An AI governance framework, policy set and accountability model10 to 14 wksFromRegister to see indicative feesISO 42001 Readiness
An audit-ready AIMS with a clause-mapped evidence pack12 to 20 wksFromRegister to see indicative feesAI Security Readiness AssessmentFLAGSHIP
A scored readiness position with a costed remediation sequence4 to 6 wksFixedRegister to see indicative feesAI Risk Assessment
A register of AI risks, rated and owner-assigned3 to 5 wksFromRegister to see indicative feesAI Governance Starter
A minimum viable policy set and an accountability map2 wksFixedRegister to see indicative feesAI Security Architecture & Threat Modelling
A threat model and a target-state architecture for AI systems6 to 10 wksFromRegister to see indicative feesAI Red Teaming & LLM Security
An adversarial findings report mapped to the OWASP LLM Top 104 to 8 wksFromRegister to see indicative feesAI Agent Security
Agent permission boundaries, controls and a monitoring design6 to 10 wksDiscoveryScopedAI Incident Response & Monitoring
AI-specific response playbooks and a detection design5 to 8 wksDiscoveryScopedAll figures AUD, excluding GST, and confirmed in writing after scoping · Full pricing detail, and what moves each figure
ONE
PROBLEM
PROBLEM
WHY THESE BELONG TOGETHERThese nine services answer one
These nine services answer one
accountability problem, not nine
technical ones.
Most organisations didn’t decide to adopt AI. It arrived: in a procurement tool, a customer service layer, a developer’s workflow, a vendor’s product update. By the time anyone asked who was accountable for it, the answer was already someone’s name on an org chart, usually without a budget line or a mandate.
What connects these nine services is not a technology. It’s a single question a board can now ask and most executives cannot answer: can you show me how AI is governed here, what it’s exposed to, and who is responsible when it goes wrong? Governance answers the first. Assessment answers the second. Architecture and testing answer the third.
We treat AI security as an accountability problem with technical consequences, not a technical problem that happens to have an accountability wrapper.
That ordering is why the governance work comes first and why the assessment is priced to be affordable before anything else is committed.
Three disciplines. One accountable system.
AI Security service categories
02
Assess
If you’ve been asked where you stand on AI risk and the honest answer is that nobody has looked, this is the group.
Usually starts with the AI Security Readiness Assessment: fixed price, four to six weeks.AI Security Readiness Assessment
A scored position across governance, exposure and controls, with a costed remediation sequence
AI Risk Assessment
Identifies and rates the specific risks your AI use creates, with owners assigned
AI Governance Starter
The two-week entry point: a minimum viable policy set and an accountability map
03
Defend & Respond
If AI systems are in production and you need to know they’ll hold under adversarial pressure, this is the group.
Usually starts with AI Security Architecture & Threat Modelling. Testing without a threat model finds the wrong things.AI Security Architecture & Threat Modelling
Designs the target-state architecture and models what an attacker would actually do
AI Red Teaming & LLM Security
Adversarial testing against the OWASP LLM Top 10 and model-specific attack classes
AI Agent Security
Permission boundaries and controls for agents that act rather than answer
AI Incident Response & Monitoring
Response playbooks and detection design for AI-specific failure modes
Not sure which one you need?
These are the situations people describe when they call. Find yours.
01
“The board asked who’s accountable for AI here and I didn’t have an answer.”
AI Governance StarterAssess · 2 weeks
02“We know people are using AI tools. We don’t know which ones, or what they’re putting into them.”
AI Security Readiness AssessmentAssess · 4 to 6 weeks
03“A customer’s procurement questionnaire asked whether we’re ISO 42001 certified, and we had to say no.”
ISO 42001 ReadinessGovern · 12 to 20 weeks
04“We’ve shipped a customer-facing AI feature and nobody has tried to break it.”
AI Red Teaming & LLM SecurityDefend & Respond · 4 to 8 weeks
05“Our agents can now take actions in live systems and I’m not sure what stops them.”
AI Agent SecurityDefend & Respond · scoped
06“We have an AI policy. Nobody follows it, and I don’t think it would survive an audit.”
AI Governance & Responsible AIGovern · 10 to 14 weeks
07“Our insurer or regulator has started asking about AI and we have nothing written down.”
AI Risk AssessmentAssess · 3 to 5 weeks
08None of these sound like us
Take the free self-assessment5 minutes · no email required
No email required · results on screen
Take the assessmentFree AI RiskSelf-Assessment
Fifteen questions across five governance domains. Returns an indicative maturity position and the two things most organisations at your level fix first. Takes about five minutes.
The order these usually happen in
Most organisations don’t buy all of these, and almost nobody buys them at once. This is the sequence we see most often.
Find out where you stand
Nothing can be sequenced until someone has looked. This is the cheapest bounded step and it prices everything after it.
Free Self-AssessmentAI Security Readiness AssessmentEstablish accountability
Governance before controls. Technical work without an owner and an approval path decays within two quarters.
AI Governance StarterAI Governance & Responsible AIDesign and test the systems
Threat modelling before red teaming. Testing without a model finds the wrong things and misses the right ones.
Architecture & Threat ModellingAI Red Teaming & LLM SecurityCertify, if you need to
Certification is a market requirement, not a security outcome. It goes last because it evidences work already done.
ISO 42001 ReadinessRuns alongside rather than after: AI Agent Security and AI Incident Response & Monitoring. Both attach to systems as they go live, not to a programme stage.
If you already have an AI governance framework and an owner, start at stage three. The sequence describes what we see most often, not a path we require. Organisations with mature cyber governance frequently skip stage two entirely, because the accountability model already exists and only needs extending.
Who leads this practice
YOUR AI SECURITY PRACTICE LEAD
Steve Sharma
Founder & Principal
Leads AI governance and AI security engagements end to end, and sits in the room for every ISO 42001 readiness programme and every red team scoping.
Full profile and published workThe standards and frameworks we work to
Where a framework has its own page, the name links to it.
03NIST AI Risk Management Framework
04OWASP LLM Top 10
05OWASP Machine Learning Top 10
06MITRE ATLAS
07EU AI Act · EU
08Australia’s AI Ethics Principles · AU
AU and EU mark frameworks that apply in a specific jurisdiction.
What this
practice doesn’t
cover
01
Operations
We are not a managed security service.
We don’t run your SOC, monitor your environment, or hold an operational pager.
02
Certification
We prepare organisations for certification.
Accredited certification bodies certify. We are not one, and we never sit on both sides of that line.
03
Model development
We don’t build, train or fine-tune models.
We govern, assess and secure the systems you or your vendors have built.
04
Referrals
If you need one of these, say so and we’ll point you to someone.
We take nothing for the referral.
How these engagements run
Scope in writing before anything starts
A fixed scope, a fixed fee or a stated range, and a named exclusions list. Nothing begins on a verbal understanding.
Evidence, not questionnaires
We inspect configurations, sample systems and test controls. A claimed control and a demonstrated one are different things and we report them differently.
Findings sequenced and costed
Every gap comes with a priority, an effort estimate and a rough cost, so the next phase can be scoped rather than guessed.
Handover that survives our departure
Documentation written for your team to run, and a named internal owner introduced before the engagement closes.
CLIENT EVIDENCE
What this work produces
Financial services400 to 800 staffAnonymised, permission on file
The challenge
A board had approved an AI programme before anyone had established who was accountable for its risk, and a customer’s procurement questionnaire arrived asking for an AI governance position the organisation did not have.
The approach
More of the workA four-week readiness assessment established the current position across governance, exposure and controls. A minimum viable policy set and an accountability model followed. The full governance framework was sequenced across two budget years rather than attempted at once.
11 weeks
From first conversation to a governance position the board signed and the customer accepted, against an internal estimate of two quarters.
What these cost
Indicative fees
Register to see indicative feesTwo services in this pillar are priced as fixed fees, five carry a “from” figure that moves with scope, and two are scoped in a discovery workshop before anything is quoted. The self-assessment is free.
What moves a “from” figure is almost always the same three things: how many systems are in scope, how much evidence already exists, and the target you’re being held to. Discovery-mode services are the ones where those variables are wide enough that quoting a number before scoping would be guessing.
Every figure, what moves it, and how engagements are contractedWho this practice is built for
A fit decision should be possible before a sales conversation.
This fits
Organisations that have already deployed AI somewhere and now need to govern it
Someone named as accountable for AI risk, with or without a budget line
Boards or executives who need a defensible position, not a technical report
Smaller organisations than our site-wide band: the Governance Starter exists for them
This doesn’t
Organisations wanting AI capability built rather than governed or secured
Anyone needing a compliance certificate issued rather than prepared for
Teams looking for a tool purchase rather than an accountability position
Work that needs ongoing operational monitoring. We don’t run a SOC
Outside that fit? Say so on the first call. We will point you toward a more appropriate firm and take nothing for the referral.
Questions about AI Security
Almost always an assessment: you can’t sequence work nobody has scoped. The free self-assessment gives an indication in five minutes; the AI Security Readiness Assessment gives an evidenced position in four to six weeks for a fixed fee. The sequence above sets out what usually follows.
All of them, but rarely at once and rarely for the same client. Most engagements start with one service and add a second after it. If you need several, the sequence matters more than the bundle, and we’d rather scope two phases well than quote a programme neither of us can predict.
Yes. The AI Security practice is jurisdiction-neutral except where a specific instrument applies: the EU AI Act work is labelled accordingly. ISO 42001, NIST AI RMF and OWASP-based work is delivered anywhere.
They are one accountability problem, not two practices. An organisation with mature cyber governance usually needs its AI work to extend that model rather than build a parallel one, and an organisation with neither should not attempt both at once. The adjacent pillar section below sets out where the boundary actually falls.
When you already know what’s wrong and need it fixed rather than found. An assessment that confirms what you already knew is a waste of your money and our time. If you have a recent, evidenced position and a clear gap list, skip the assessment and come to us with the gap list. We’ll tell you if we think it needs re-testing, and we’ll say so if we don’t.
Then start with the free self-assessment. It costs nothing, asks for no email, and returns an indicative position in five minutes. It is genuinely the right first step for a large share of the people who read this page, and a great many of them never need to buy anything from us at all.
THE ADJACENT PILLAR
Cybersecurity: 12 services If your question is actually about cybersecurity
AI security and cybersecurity are one accountability problem, not two practices, but they are bought separately, and a reader who arrived here because AI is the visible symptom sometimes needs the foundational work first. If what you actually need is ISO 27001, an Essential Eight position or a security strategy, that’s the other pillar.
CHOOSE YOUR NEXT STEP
Most people who read this page start with one of three things.
Indicative fees are shown to registered users and confirmed in writing after scoping.