Submission 75
Tata Consultancy Services
Submitter group: Companies.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
TCS, which describes its own data-centre and sovereign-cloud offerings, argues Australia needs sufficient AI access, choice and control rather than every layer of the stack. It recommends strategic compute infrastructure, open architecture, accountable governance and skills, reserving sovereign infrastructure for workloads with clear public benefit and weighing its costs against continuity and control.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Argues that reserving access to foreign-owned compute does not create Australian capability, as infrastructure, models and decisions remain under foreign control.
However, reserving access to externally owned compute does not create Australian capability. It may ease short-term constraints but the underlying infrastructure, models, commercial terms and strategic decisions remain under foreign control.
States that Australia does not need to build every layer of the AI stack but needs sufficient access, choice and control.
Australia does not need to build every layer of the AI stack. We need sufficient access, choice and control to safeguard our interests and reduce exposure to disruptions to critical technology supply chains caused by geopolitical tension and trade restrictions.
Describes data sovereignty through location, governance and law; technical sovereignty through infrastructure and model control; and operational sovereignty through people, processes and accountability.
1. Data sovereignty – where data is held, how it is governed and which laws may apply. 2. Technical sovereignty – control over infrastructure, models, interfaces and architecture. 3. Operational sovereignty – the people, processes and accountability needed to operate, secure, maintain and adapt the capability.
Argues that smaller or locally hosted models fine-tuned on Australian datasets can serve many routine tasks without frontier models.
Many routine tasks do not require frontier models. Smaller or locally hosted models that are fine-tuned with Australian datasets can reduce operating costs, provide greater control over sensitive data and operations, and deliver better outcomes.
Argues that skills are central to operational sovereignty and economic growth, since infrastructure and models cannot deliver sovereign capability without people to operate them.
Infrastructure and models cannot deliver sovereign capability without the people to operate, secure and adapt them. Skills are central to operational sovereignty and economic growth.
Argues that governance should be continuous, with full lineage across data, model, decision and outcome and visible oversight and audit evidence.
Governance should be continuous rather than periodic, with full lineage maintained across data, model, decision and outcome, so that oversight is visible and evidence is available for audit.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Assess the business case for investing in a sovereign AI capability.
Governments should assess the business case for investing in such a sovereign capability.
Treat AI compute as strategic national infrastructure, assessed for resilience, continuity and geopolitical exposure, not solely on cost or availability.
Treat AI compute as strategic national infrastructure, assessed for resilience, continuity and geopolitical exposure — not solely on cost or availability.
Build sovereign capability in addition to reserving foreign compute, since access to externally owned infrastructure does not equate to Australian agency.
Build sovereign capability in addition to reserving foreign compute, recognising that access to externally owned infrastructure does not equate to Australian agency.
Mandate open architecture, interoperability and portability in publicly funded AI capability to preserve future choice and avoid single-provider dependence.
Mandate open architecture, interoperability and portability in publicly funded AI capability to preserve future choice and avoid single-provider dependence.
Embed a responsible AI assurance and accountability framework across the AI lifecycle, with clear ownership, human oversight and auditable evidence.
Embed a responsible AI assurance and accountability framework across the AI lifecycle, with clear ownership, human oversight and auditable evidence.
Invest in Australian operational, cybersecurity and data governance skills as a core element of sovereignty and economic diversification.
Invest in Australian operational, cybersecurity and data governance skills as a core element of sovereignty and economic diversification.
Connect infrastructure investment to research commercialisation and industry so value is captured through Australian IP, products and services, not data centres alone.
Connect infrastructure investment to research commercialisation and industry, so that value is captured through Australian intellectual property, products and services — not data centres alone.
Facilitate access to advanced AI compute for government agencies, researchers and start-ups, as governments have done with supercomputer infrastructure.
Just as governments have previously developed and maintained supercomputer infrastructure, they should facilitate access to advanced AI compute for government agencies, researchers and start-ups.
For critical government workloads, assess whether geopolitical exposure, foreign restrictions or supplier decisions could interrupt access.
For critical workloads, that assessment should also consider geopolitical exposure, including whether foreign restrictions or supplier decisions could interrupt access.
Reserve sovereign infrastructure for workloads where greater Australian control produces a clear public benefit.
Sovereign infrastructure should be reserved for workloads where greater Australian control produces a clear public benefit.
Support a mix of closed and open-weight models matched to the task in sovereign capability architecture.
The architecture should support a mix of closed and open- weight models matched to the task.
Weigh sovereign capability's upfront investment, deployment time and skilled workforce needs against continuity, control, portability and reduced long-term provider dependence.
Establishing this capability involves trade-offs. It requires more upfront investment, takes longer to deploy and needs a skilled workforce to operate and maintain the system. These costs should be weighed against the value of continuity, control, portability and reduced exposure to long-term provider dependence.
Use strategic oversight for direction, risk appetite and higher-risk approvals; standing governance and assurance for standards, use cases, controls, value and compliance; and operational teams to build, monitor and manage systems.
A workable model operates across three connected tiers: • Strategic oversight — senior sponsorship that sets direction, risk appetite and approvals for higher-risk uses. • Governance and assurance — a standing function that maintains standards, reviews use cases, oversees responsible-AI controls, and tracks value and compliance. • Operational delivery — the teams that build, run and monitor AI systems, implement controls, and manage models and agents day to day.
Identify an accountable owner for AI use and assurance sign-off, with continuous governance and auditable lineage across data, models, decisions and outcomes.
Underpinning these tiers, a clearly identified accountable owner should hold responsibility for AI use and assurance sign-off. Governance should be continuous rather than periodic, with full lineage maintained across data, model, decision and outcome, so that oversight is visible and evidence is available for audit.
Build skills across specialist system and model designers, practitioners integrating AI into services, policy and operations, and whole-of-workforce AI literacy.
Capability should be built across three broad tiers: • A specialist cadre who design, build and fine-tune AI systems and models. • A wider group of practitioners who build with AI — integrating and applying it within services, policy and operations. • A whole-of-workforce baseline of AI literacy, so that the majority who use AI in daily work can do so effectively and responsibly.
Embed security and reliability, human accountability and intervention, lifecycle bias monitoring, explainability, privacy and identity protection in responsible sovereign AI.
Responsible AI should therefore be treated as a core dimension of sovereignty, resting on a consistent set of principles: • Security and reliability — protection against misuse, data leakage and adversarial attack, with resilience and continuity built in. • Accountability — clear ownership of AI outcomes, with defined human oversight and the ability to intervene or override. • Fairness — active management of bias, with monitoring across the lifecycle rather than one-off testing at deployment. • Transparency and explainability — decisions that can be interpreted, audited and explained to those affected. • Privacy and identity protection — safeguards for personal and sensitive information across data, models and operations.
Embed responsible AI principles throughout the lifecycle with guardrails and auditable evidence; extend assurance to agentic systems' actions as well as outputs.
These principles are most effective when embedded across the full AI lifecycle — from use-case selection and data preparation through to deployment and ongoing monitoring —supported by guardrails and auditable evidence. As AI systems become more autonomous and agentic, this assurance must extend to how systems act, not just their output.