Submission 552
Thomson Reuters
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.
A professional information provider argues Australia should measure how well, not only how quickly, AI is adopted. It proposes a National Trusted AI Benchmark, risk-based data sovereignty, workforce training, controlled Commonwealth pilots, licensed copyright use and strengthened existing national-security arrangements. It says it sells AI products to government and proposes its own pilot and Foundation-based benchmark.
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.
Thomson Reuters argues Australia should assess the quality of AI adoption, not only its speed or uptake.
Australia should measure not only how quickly AI is adopted, but how well it is adopted.
Thomson Reuters says the proposed initiative would provide annual evidence to target skills programs, implementation support and regulatory attention, including SME adoption barriers.
Importantly, this would give the Australian Government an annual evidence base to target skills programs, implementation support and regulatory attention, including where small and medium businesses face adoption barriers.
Thomson Reuters argues that for critical infrastructure and high-consequence uses, government and operators should understand where data is stored, which jurisdiction applies and who can access it.
For critical infrastructure and other high- consequence uses, government and operators should understand where sensitive data is stored and processed, which jurisdiction applies, who can access it, how it may be reused, and whether access can be maintained during disruption or a change of supplier.
Thomson Reuters argues the workforce challenge includes whether AI weakens the development of professional judgement and whether workers get the training to use it responsibly.
The workforce challenge is not only job creation or displacement. It is whether AI weakens the development of professional judgement and whether workers receive the training needed to use it responsibly.
Thomson Reuters recommends Commonwealth AI use in legal, policy, tax and other high-volume functions be piloted under controlled conditions before scaling.
Use controlled Commonwealth pilots in legal, policy, tax and other high-volume functions before scaling. Pilots should use authoritative information, defined human oversight, privacy and security controls, auditability and published evaluation criteria.
A program corrected obvious text-recognition errors in this quote for display.
Thomson Reuters argues copyright policy should support both AI innovation and sustainable content creation, with licensing markets providing lawful access to quality material.
Copyright policy should therefore support both AI innovation and sustainable content creation. These objectives are not mutually exclusive. Well-designed licensing markets can provide AI developers with lawful access to high-quality material while preserving incentives for Australian creators, publishers and news organisations to continue investing in original work.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Establish a National Trusted AI Benchmark to measure the quality and outcomes of AI adoption, including productivity, accuracy, workforce capability, trust and accountability, rather than uptake alone.
ToR A & D: Establish a National Trusted AI Benchmark to measure the quality and outcomes of AI adoption, including productivity, accuracy, workforce capability, trust and accountability, rather than uptake alone.
Base the National Trusted AI Benchmark on an Australian deployment of the Thomson Reuters Foundation and UNESCO's AI Company Data Initiative.
Establish a National Trusted AI Benchmark based on an Australian deployment of the AI Company Data Initiative, developed by the Thomson Reuters Foundation with UNESCO.
Measure governance, oversight, workforce consultation, skills investment, job quality, copyright and creator rights, SME capability, incidents and productivity-benefit distribution; add Australian modules on consultation, consumer risk, procurement and SME participation.
The benchmark should measure governance maturity, human oversight, workforce consultation, skills investment, job quality, copyright and creator rights, SME capability, incident reporting and the distribution of productivity benefits. Australian modules should cover workplace consultation, consumer risk, government procurement and SME participation.
A program corrected obvious text-recognition errors in this quote for display.
Adopt risk-based sovereignty, identifying critical data, compute, systems and skills; require appropriate Australian residency, jurisdiction, access and continuity controls for high-consequence uses, rather than technological self-sufficiency.
ToR E & F: Adopt a risk-based approach to sovereign AI and data sovereignty. Identify the data, compute, systems and skills that are critical to national security, economic continuity and effective government, and require appropriate Australian residency, jurisdiction, access and continuity controls for high-consequence uses. Sovereignty should mean control where it matters, not technological self-sufficiency.
A program corrected obvious text-recognition errors in this quote for display.
Apply risk-based data sovereignty requirements to high-consequence AI, with clear expectations for where sensitive data is stored and processed, who can access it, applicable jurisdiction, and continuity of access during disruption or supplier change.
Applying risk-based data sovereignty requirements to high-consequence AI, including clear expectations for where sensitive data is stored and processed, who can access it, the jurisdiction that applies, and continuity of access during disruption or supplier change.
Prioritise trusted, domain-specific AI for high-consequence uses, grounded in authoritative and licensed information, with traceable sources and clear human accountability.
Prioritising trusted, domain-specific AI for high-consequence uses, grounded in authoritative and licensed information, with traceable sources and clear human accountability.
A program corrected obvious text-recognition errors in this quote for display.
Support trusted, domain-specific AI grounded in authoritative Australian data where appropriate.
Support trusted, domain- stored, processed and accessed. specific AI grounded in authoritative Australian data where appropriate.
A program corrected obvious text-recognition errors in this quote for display.
Strengthen vendor and supply-chain assurance, including assessment of data flows, ownership, model and compute dependencies, security controls and continuity risks.
Strengthening vendor and supply-chain assurance, including assessment of data flows, ownership, model and compute dependencies, security controls and continuity risks.
A program corrected obvious text-recognition errors in this quote for display.
Require testing, auditability and human oversight before AI is deployed at scale, building on existing ACSC guidance and using measurable assurance rather than high-level principles alone.
Require testing, auditability and human oversight before AI is deployed at scale, building existing ACSC guidance and using measurable assurance rather than relying on high-level principles alone.
Address vendor and supply-chain dependencies, data provenance and security, model testing, human oversight, escalation and incident response in governance.
Governance should also address vendor and supply-chain dependencies, the provenance and security of data, model testing, human oversight, escalation and incident response.
Make workforce capability a condition of responsible AI adoption, supported by role-specific training, supervised practice, verification, mentoring and clear human accountability.
ToR G: Make workforce capability a condition of responsible AI adoption, supported by role-specific training, supervised practice, verification, mentoring and clear human accountability.
A program corrected obvious text-recognition errors in this quote for display.
Have government, education providers and employers fund role-specific training, supervised review, exposure to errors, mentoring and clear rules on when human expertise must override the technology, with workforce measures part of AI procurement and evaluation.
Government, education providers and employers should fund role-specific training, supervised review of AI- assisted work, exposure to errors and edge cases, mentoring from experienced practitioners and clear rules on when human expertise must override the technology. Workforce measures should form part of AI procurement and evaluation, not be considered after deployment.
A program corrected obvious text-recognition errors in this quote for display.
Use controlled Commonwealth pilots in legal, policy, tax and other high-volume functions before scaling, with authoritative information, defined human oversight, privacy and security controls, auditability and published evaluation criteria.
ToR H & I: Use controlled Commonwealth pilots in legal, policy, tax and other high-volume functions before scaling. Pilots should use authoritative information, defined human oversight, privacy and security controls, auditability and published evaluation criteria.
A program corrected obvious text-recognition errors in this quote for display.
Proposes a 12-week CoCounsel pilot through normal procurement for legal and policy research, regulatory comparison, document review, summarisation and first-draft briefs, excluding automated determinations affecting rights or entitlements.
Through its AI Policy White Paper, Thomson Reuters has proposed a 12-week CoCounsel pilot to the Government, through normal procurement processes for legal and policy research, regulatory comparison, document review, summarisation and first-draft briefs. Automated determinations affecting rights or entitlements would be excluded.
A program corrected obvious text-recognition errors in this quote for display.
Require authorised data, licensed content, authoritative sources and citations, named responsibility, verification, auditability, tiered review empowered to pause or alter use, training, material-use disclosure, escalation and recourse.
Guardrails should include authorised data and licensed content, authoritative sources and citations, named human responsibility, mandatory verification, auditability, tiered review with authority to pause or alter use, role- specific training, disclosure where AI materially shapes an output, and clear escalation and recourse.
A program corrected obvious text-recognition errors in this quote for display.
Evaluate legal and policy pilots on time saved, factual and citation accuracy, contestability, privacy, security, workforce confidence and effects on professional capability.
Evaluation should measure time saved alongside factual and citation accuracy, contestability, privacy, security, workforce confidence and effects on professional capability.
A program corrected obvious text-recognition errors in this quote for display.
Test AI-enabled anomaly detection, reconciliation and compliance in controlled Commonwealth use cases.
Test AI-enabled anomaly detection, reconciliation and compliance in controlled Commonwealth use cases.
Consider measuring administrative cost, processing time, accuracy, false positives, attributable additional revenue or avoided leakage, contestability, workforce impacts and auditability in Commonwealth tax pilots.
Commonwealth pilots could measure administrative cost, processing time, accuracy, false-positive rates, additional revenue or avoided leakage where attributable, contestability, workforce impacts and auditability.
Apply risk-tiered governance to high-consequence AI and preserve the value of protected human-created content.
ToR I & J: Apply risk-tiered governance to high-consequence AI and preserve the value of protected human- created content.
Ensure copyrighted material used in AI development is authorised, appropriately attributed and fairly compensated, supported by practical licensing arrangements and transparency about provenance and use.
Copyrighted material used in AI development should be authorised, appropriately attributed and fairly compensated, supported by practical licensing arrangements and transparency about provenance and use.
Support a copyright framework protecting human-created content in AI development and deployment and practical licensing for lawful commercial access to high-quality Australian content.
Thomson Reuters recommends that the Committee support a copyright and AI framework that: • preserves the application of copyright to protect human-created content used in AI development and deployment; • supports practical licensing mechanisms that allow AI developers to access high-quality Australian content on lawful commercial terms;
Require attribution when protected content is reproduced, surfaced or materially relied upon; transparency about provenance and use; and human accountability, verification and governance for high-consequence AI.
requires appropriate attribution where protected content is reproduced, surfaced or materially relied upon; • promotes transparency in the provenance and use of protected content in AI systems; and • ensures that high-consequence uses of AI remain subject to human accountability, verification and governance.
A program corrected obvious text-recognition errors in this quote for display.
Strengthen existing national-security AI links; implement ACSC guidance through testing and assurance; scrutinise sensitive data, vendors, supply chains, research-security risks and strategic dependencies.
ToR L & M: Build on, rather than duplicate, Australia’s existing national-security and critical-infrastructure AI arrangements. Strengthen operational links between Home Affairs, Defence, the National Intelligence Community, law enforcement, AISI, regulators and trusted industry; implement existing ACSC guidance through testing and assurance; and strengthen scrutiny of sensitive data, vendors, technology supply chains, research-security risks and strategic dependencies.
A program corrected obvious text-recognition errors in this quote for display.
Formalise operational links among Home Affairs, the National Intelligence Community, law enforcement, Defence, AISI, regulators and trusted industry.
Formalise operational links between Home Affairs, the National Intelligence Community, law enforcement, Defence, AISI, relevant regulators and trusted industry capability.
A program corrected obvious text-recognition errors in this quote for display.
Support AI analysis of cyber threats, interference, research-security risks, unwanted technology transfer, supply chains and transnational networks; securely use relevant government, open-source and licensed data; monitor continuously, with human responsibility.
Existing arrangements should support AI-enabled analysis of cyber threats, foreign interference, research-security risks, unwanted technology transfer, supply-chain dependencies and transnational networks; secure use of relevant government, open-source and appropriately licensed commercial data; continuous rather than point-in-time risk monitoring; and clear human responsibility for consequential decisions.
Give AISI structured information-sharing links with national-security agencies so technical findings on emerging AI capabilities can be considered alongside operational intelligence.
AISI should have structured information-sharing links with national-security agencies so technical findings about emerging AI capabilities can be considered alongside operational intelligence.
A program corrected obvious text-recognition errors in this quote for display.
Use existing Home Affairs technology-security and foreign-interference mechanisms where possible rather than duplicating them.
Existing Home Affairs mechanisms, including its technology-security and foreign-interference work, should be used where possible rather than duplicated.
A program corrected obvious text-recognition errors in this quote for display.
Have the Office of AI, in the first 30 days of a proposed sequence, agree the problems to be tested, limits of use, governance, workforce safeguards and evaluation criteria.
Thomson Reuters has developed an AI policy white paper that proposes a practical sequence that keeps policy control with government while building evidence before scale. In the first 30 days, the Office of AI would agree the problems to be tested, limits of use, governance, workforce safeguards and evaluation criteria.
A program corrected obvious text-recognition errors in this quote for display.
Have government, by day 60 of the proposed sequence, select benchmark scope and pilot teams, complete security, privacy and workforce consultation, establish accountable human owners and escalation arrangements, and constitute a multi-stakeholder advisory group.
By day 60, government would select benchmark scope and any pilot teams, complete security, privacy and workforce consultation, establish accountable human owners and escalation arrangements, and constitute a multi- stakeholder advisory group.
By day 90 of the proposed sequence, start the Australian benchmark and any commissioned pilots against a published evaluation framework, sharing de-identified lessons on productivity, job quality, professional capability, rights and public confidence.
By day 90, the Australian benchmark and any commissioned pilots would commence against a published evaluation framework, with a process for sharing de-identified lessons on productivity, job quality, professional capability, rights and public confidence.
A program corrected obvious text-recognition errors in this quote for display.
Read the original
In some quotes a program corrected obvious text-recognition errors for display.