You're offline. Anything you type is saved on this device and you can send it when you're back online.
Our approach
Fenced AI: the other 90% of the job
A model that gives the right answer is only the beginning. Fenced AI is the Strata Model applied to AI: we wrap every capability in the layers that make it trustworthy in production.
Explore the six layers
We surround every AI capability with the layers that make it safe to run: architecture, governance, security, content safety and infrastructure fit. Select a ring to explore.
Select a ring to explore
Governance
Ownership, risk, oversight and regulation.
Who owns the system, how risk is classified, who can approve changes and how the system aligns with local AI and data-protection rules.
Questions it answers
- • Who is accountable for outcomes?
- • What risk class does this use case fall into?
- • How do we evidence compliance to a regulator?
Typical deliverables
- Governance framework and RACI
- Risk classification and register
- Model and policy documentation
Governance
Ownership, risk, oversight and regulation.
Who owns the system, how risk is classified, who can approve changes and how the system aligns with local AI and data-protection rules.
Questions it answers
- • Who is accountable for outcomes?
- • What risk class does this use case fall into?
- • How do we evidence compliance to a regulator?
Typical deliverables
- Governance framework and RACI
- Risk classification and register
- Model and policy documentation
What each layer covers
01
Capability
The AI capability at the centre: a model, an agent, a retrieval system or an analytics engine, chosen for the problem rather than the hype.
Questions it answers
- Which problem is worth solving with AI?
- Which model or approach fits the data we have?
- How will we measure if it works?
Typical deliverables
- Use-case definition
- Model selection rationale
- Evaluation plan and metrics
02
Architecture
How AI components connect to data, core systems and users, designed for local infrastructure and the channels people actually use.
Questions it answers
- Where does the data come from and how fresh is it?
- How does this integrate with core systems?
- What happens when the network is slow or down?
Typical deliverables
- Reference and solution architecture
- Integration and data-flow design
- Resilience and fallback patterns
03
Governance
Who owns the system, how risk is classified, who can approve changes and how the system aligns with local AI and data-protection rules.
Questions it answers
- Who is accountable for outcomes?
- What risk class does this use case fall into?
- How do we evidence compliance to a regulator?
Typical deliverables
- Governance framework and RACI
- Risk classification and register
- Model and policy documentation
04
Security
Protection of models, data and pipelines against AI-specific threats such as prompt injection, data leakage, model theft and poisoning.
Questions it answers
- Can a user trick the system into leaking data?
- Who can access the model and its training data?
- How would we detect an attack?
Typical deliverables
- AI threat model
- Security controls and hardening
- Red-team and test results
05
Content Safety
Controls on what the system produces and how it behaves: harmful content, hallucination, tone and cultural appropriateness across languages.
Questions it answers
- What should the system never say or do?
- How do we catch wrong or harmful answers?
- Does it behave well in every language we support?
Typical deliverables
- Content safety policy
- Guardrails and filters
- Multilingual evaluation set
06
Infrastructure Fit
Compute, connectivity and data residency that work in-country, with sensible alternatives where infrastructure falls short.
Questions it answers
- Can this run on the compute and bandwidth available?
- Does data need to stay in-country?
- What does it cost to run at scale?
Typical deliverables
- Hosting and residency options
- Capacity and cost model
- Edge and offline strategy
Engagement lifecycle
- 1
Discover
Goals, constraints, stakeholders and context.
- 2
Assess
Readiness and risk across every layer.
- 3
Design
Architecture, controls and plan.
- 4
Guide delivery
Build with you or oversee partners.
- 5
Sustain
Monitor, improve and transfer skills.
The “Did you consider…?” checklist
Blind spots we see again and again. If any of these make you pause, it's worth a conversation.
Want a second pair of eyes on your AI system?
Our AI Readiness & Risk Review checks every layer in three to four weeks.