In summary
- Airservices Australia has issued an RFI for an Enterprise AI Broker Platform to serve as a single, governed entry point for all AI use within its private AWS tenancy.
- The platform must support provider-independent multi-model orchestration, alongside governance, security and IRAP PROTECTED compliance controls.
- Beyond enterprise capabilities like RAG and governed connectors, the platform must also handle airspace and aeronautical spatial data.
Airservices Australia has gone to the market for information on an Enterprise AI Broker Platform - a single, governed entry point for all AI use across the organisation.
The request for information (RFI) is being issued, Airservices says, to identify market capabilities, commercial models and potential partners for the design, build, deployment and ongoing managed-service operation of the platform within the authority’s private Amazon Web Services (AWS) tenancy.
Airservices - the Commonwealth statutory authority responsible for air navigation services across Australian-administered airspace - says that, as part of its digital transformation, it wants to consolidate and govern AI use across its workforce behind a single enterprise-grade platform rather than allow a proliferation of unmanaged point solutions.
It frames the broker as the single governed entry point for all AI interactions, intended to preserve data sovereignty, compliance, security, cost control and operational visibility “without constraining the utility available to staff.”
In scope, the platform is described as a centrally governed AI workspace that consolidates conversational AI, knowledge retrieval, workflow automation, tool integration and operational observability into one platform - the organisation’s “AI front door,” replacing fragmented, unmanaged AI tool usage.
It must support multi-model orchestration, routing each request to the most suitable foundation model, integrate with corporate systems through governed connectors, and give administrators full visibility of usage, cost, behaviour and compliance.
The RFI sets out a dozen core capability areas.
Alongside a conversational interface with document analysis, voice input and in-canvas collaborative authoring, it calls for retrieval-augmented generation grounded in organisational document stores of up to 10,000 controlled documents with source citations; governed connectors to corporate systems such as Microsoft 365, SharePoint, Teams, Jira, ServiceNow and SAP; workflow automation with reusable “skills”; semi- and fully autonomous agents bounded by human-in-the-loop guardrails; and the low-code generation of interactive AI apps such as dashboards and map-based consoles.
On model choice, Airservices is insisting on provider independence.
The RFI names Anthropic, OpenAI, AWS Bedrock, Google, Mistral and Cohere as example providers, requires intelligent routing by task type, cost and latency, and stipulates that substituting a model, provider or model-serving layer must be achievable by configuration rather than redevelopment - with routing logic, prompts and guardrails kept portable and readable independently of any one provider’s proprietary tooling.
The platform must enforce role-based access control and data sovereignty - all model calls routed through Airservices-controlled cloud accounts, with no data leaving approved boundaries - and comply with the Australian Government’s IRAP PROTECTED framework.
It also calls for content filtering of sensitive and classified data, hallucination detection with source-grounding and confidence indicators, and controls for jailbreak and prompt-injection attempts, with respondents asked to state whether those controls sit in the broker itself, in the model-serving layer such as an AWS Bedrock AgentCore harness, or both.
Comprehensive audit logging, per-user and per-model cost tracking with chargeback, centralised administration and real-time collaborative workspaces round out the operational requirements.
Airservices also wants the platform to ingest and interpret airspace, surveillance-coverage, flight-path and noise-monitoring data, answer natural-language spatial queries and render results on maps, connect to Esri ArcGIS Enterprise and OGC services, and handle aeronautical data formats such as AIXM and datums including GDA2020 - with a firm caveat that outputs are for analysis and decision support only and must not be represented as certified aeronautical information.
The RFI closes at 10:00am ACT local time on 30 September 2026.




