Protect your data, control how AI uses it
The ten guardrails of the Australian Voluntary AI Safety Standard, operated 24/7 in your AWS accounts. Data protected at ingestion, access restricted to sensitive information, every AI action logged for safety review. The managed enforcement layer of the AI Factory.
Your engineers build the product. We operate the safety floor.
What you get
Data governance and security
Classification, encryption and provenance tracking enforced at ingestion. Cybersecurity controls applied to every AI data flow.
Model data isolation
Guardrails control which data can reach which model, under which identity and with what evidence. Bedrock helps remove the public model training path; we still operate the internal AWS controls it does not replace.
Accountability process
Clear ownership, internal capability and safety decision rights. Training programs for teams that build and operate agents.
Risk management
Stakeholder-impact assessments and ongoing risk reviews. Mitigations mapped to system usage and updated continuously.
Human control and oversight
Intervention points across the AI lifecycle. Meaningful human review for decisions that matter. Documented escalation paths.
User disclosure and challenge
End users told when they are talking to AI. Clear paths to challenge a decision, outcome or interaction.
Record keeping and testing
Comprehensive AI inventory. Pre-deployment testing and continuous monitoring for behaviour change or unintended consequences.
One fixed monthly fee, everything included
Operations
- 24/7 monitoring of guardrail enforcement
- Continuous safety-policy patching
- Policy drift detection and remediation
- Incident response for unsafe AI behaviour
- Monthly safety posture review
Engineering
- Core ten guardrails deployed on day one, with 60+ total available
- Custom safety-rule authoring and review
- Safety telemetry schema and event pipelines
- Data-classification tooling
- Safe tool-use and action boundaries
Assurance
- Human-review and escalation records
- User disclosure and challenge records
- Supply-chain and model transparency records
- Stakeholder-impact assessment history
- Safety review evidence when your team needs it
How it works
Guardrails are the first service deployed. Agentic Ops and MLOps inherit them.
Assess
Your AI use case, data sensitivity and current safety controls. We map the risk to the core ten and prioritise the rest of the 60+.
Build
Platform controls deployed. Safety policies authored in Git. Telemetry and review records stood up in your AWS accounts.
Operate
24/7 enforcement monitoring. AI actions logged. Safety violations surfaced before they become incidents.
Improve
New model risks and regulatory expectations tracked. Policies iterated with your product team. Safety rules rolled forward.
Mapped to the Voluntary AI Safety Standard
We treat the ten areas in the Australian Voluntary AI Safety Standard as safety categories. The actual guardrails are the policies, monitors and review records we operate beneath them.
Governance and risk
- Accountability process
- Risk management process
- Data governance, privacy and cyber security
- Testing and ongoing monitoring
Human oversight
- Human control and intervention
- Disclosure of AI use to end users
- Challenge paths for AI-affected decisions
Transparency and records
- Supply chain transparency
- Records to support assessment
- Stakeholder engagement
See where you stand with our free self-service KickOff assessment.
Audited and certified
AWS Advanced Partner
ISO 27001 Certified
AWS SaaS Competency
Chat with us about your AI project
We will show you how to make it secure, governed and successful.
Frequently asked questions
What are AI Guardrails?
AI Guardrails are managed controls that govern how agents and models use data, tools and actions in production, with access boundaries, review paths, safety telemetry and evidence.
Are guardrails a separate product?
No. Guardrails are part of the AI Factory. They make AI safer in production by protecting data, limiting unsafe actions, controlling tool access and creating review paths.
What AI safety risks do the guardrails cover?
Sensitive data exposure, unintended data use, unsafe agent actions, prompt or tool misuse, missing human oversight, poor disclosure and weak traceability.
How are the guardrails enforced?
Through platform controls, not policy documents. We apply data classification, access boundaries, tool-use limits, human review triggers and safety monitoring.
Can we add our own safety rules?
Yes. Product-specific safety rules sit on top of the baseline, including approved actions, blocked topics, human approval points and escalation paths.
Do guardrails replace human oversight?
No. They define where AI can act on its own, where it must ask for approval and where activity should be escalated for review.
What records do the guardrails keep?
Safety telemetry: what data was accessed, which tools were used, which policy applied, whether a human reviewed the action and what happened next.