AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 159

Trustethica

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.

Open in the interactive guide

Summary

Read the original submission for the full text.

Trustethica, which provides AI governance infrastructure for enterprises, argues that consequential AI needs operational, runtime governance rather than principles and pre-deployment assessment alone. It proposes a nationally consistent, risk-proportionate assurance framework, accountability records, runtime expectations for high-consequence and agentic AI, funded pilots, and an Australian profile aligned to ISO/IEC 42001.

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.

  1. Argues that governance of consequential AI must continue while systems operate, not end at a policy statement, register or pre-deployment assessment.

    For consequential AI—particularly systems used in regulated, high-consequence or public-facing contexts—governance cannot end with a policy statement, an AI register or a pre-deployment assessment. Governance must continue while systems operate and encounter changing data, users, contexts, capabilities, suppliers and incentives.

  2. Recommends that organisations keep a structured accountability record for consequential AI systems.

    For consequential systems, organisations should maintain a structured accountability record.

  3. Argues that governance supports adoption because uncertainty over accountability, required controls and evidence can delay organisations’ valuable AI deployment.

    Responsible governance is adoption infrastructure, not merely a compliance cost. Organisations often delay valuable AI deployment because they cannot determine who is accountable, what controls are required, whether suppliers are acceptable, how privacy and security obligations apply, or what evidence will be needed if problems arise.

  4. Argues that governance infrastructure is part of sovereign capability, including understanding where data is used and who can access it.

    Governance infrastructure is therefore part of sovereign capability. Australian organisations should be able to understand and evidence where their data is used, who can access it, which systems and suppliers influence an AI outcome, which obligations apply, and which accountable person can halt or constrain use.

  5. Argues that AI used by agencies that affects rights, entitlements or access to services needs a particularly high standard of transparency, contestability and accountability.

    Commonwealth agencies can use AI to improve accessibility, service quality, timeliness, administrative efficiency and staff capability. However, systems that affect rights, entitlements or access to services require a particularly high standard of transparency, contestability and accountability.

  6. Recommends stronger expectations for agentic AI on authority boundaries, tool access, logging and revocation.

    For agentic AI, Australia should set stronger expectations around explicit authority boundaries, tool access, approval gates, segregation of duties, action logging, revocation mechanisms, fail-safe behaviour and review of unexpected or high-impact actions.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Develop a nationally consistent, technology-neutral AI assurance framework distinguishing low-risk from consequential/high-risk use, with baseline accountability, recordkeeping, risk-assessment and incident-management expectations and enhanced requirements where harms are substantial, scalable or hard to reverse.

    Establish a nationally consistent, risk-proportionate AI assurance framework. The Australian Government should develop a technology-neutral framework that differentiates low- risk AI use from consequential and high-risk use. It should establish baseline expectations for accountability, recordkeeping, risk assessment and incident management, with enhanced requirements where potential harms are substantial, scalable or difficult to reverse.

  2. Require proportionate records for consequential AI covering purpose, boundaries, ownership, affected persons, dependencies, obligations, risk classification, controls, monitoring, significant changes and residual-risk decisions.

    Require purpose, ownership and evidence for consequential AI. Organisations operating consequential AI should maintain a proportionate accountability record covering authorised purpose and boundaries, accountable ownership, affected persons, material dependencies, applicable obligations, risk classification, controls, monitoring, significant changes and residual-risk decisions. Material expansion of scope, autonomy, audience, data use or decision authority should trigger reassessment and appropriate approval.

  3. Trigger reassessment and appropriate approval when consequential AI materially expands its scope, autonomy, audience, data use or decision authority.

    Require purpose, ownership and evidence for consequential AI. Organisations operating consequential AI should maintain a proportionate accountability record covering authorised purpose and boundaries, accountable ownership, affected persons, material dependencies, applicable obligations, risk classification, controls, monitoring, significant changes and residual-risk decisions. Material expansion of scope, autonomy, audience, data use or decision authority should trigger reassessment and appropriate approval.

  4. Include intended users, human authority, testing and assurance evidence, escalation thresholds, material incidents, residual-risk acceptances and review or decommissioning criteria in consequential systems’ accountability records, proportionate to risk.

    For consequential systems, organisations should maintain a structured accountability record. Proportionate to the risk, this record should identify the authorised purpose and limitations; intended users and affected persons; accountable owners; the system’s role and degree of human authority; material model, data, tool, vendor and infrastructure dependencies; relevant legal and policy obligations; testing and assurance evidence; monitoring indicators and escalation thresholds; material incidents and changes; residual-risk acceptances; and review or decommissioning criteria.

  5. Set minimum runtime-governance expectations for high-consequence and agentic AI: ongoing monitoring against defined indicators/thresholds, traceable escalation, controlled exception handling, action/decision logging, human intervention and evidence preservation.

    Set minimum runtime-governance expectations for high-consequence and agentic AI. Significant systems should be subject to ongoing monitoring against defined indicators and thresholds, traceable escalation, controlled exception handling, action and decision logging, human-intervention mechanisms and preservation of evidence. Agentic systems should additionally be subject to explicit controls over authority, tool access, approvals, traceability, segregation of duties, revocation and safe failure.

  6. For agentic systems, additionally control authority, tool access, approvals, traceability, segregation of duties, revocation and safe failure.

    Set minimum runtime-governance expectations for high-consequence and agentic AI. Significant systems should be subject to ongoing monitoring against defined indicators and thresholds, traceable escalation, controlled exception handling, action and decision logging, human-intervention mechanisms and preservation of evidence. Agentic systems should additionally be subject to explicit controls over authority, tool access, approvals, traceability, segregation of duties, revocation and safe failure.

  7. Set stronger expectations for review of unexpected or high-impact actions by agentic AI.

    For agentic AI, Australia should set stronger expectations around explicit authority boundaries, tool access, approval gates, segregation of duties, action logging, revocation mechanisms, fail-safe behaviour and review of unexpected or high-impact actions.

  8. Fund time-bounded pilots in Commonwealth services and regulated or nationally significant sectors, evaluated on control effectiveness, oversight, impacts, incident handling, cost, interoperability and evidence quality as well as performance.

    Government should support time-bounded pilots in Commonwealth services and regulated or nationally significant sectors. Evaluation should address not only technical performance and productivity, but also control effectiveness, human oversight, impacts on affected people, incident handling, implementation cost, interoperability and assurance-evidence quality.

  9. Where safe and appropriate, publish de-identified lessons, reusable control patterns and procurement guidance from the pilots.

    Where safe and appropriate, government should publish de-identified lessons, reusable control patterns and procurement guidance.

  10. Government, regulators, standards bodies, industry, civil society and technical experts should develop an Australian profile aligned with international standards, including ISO/IEC 42001, Australian law and sectoral requirements.

    Government, regulators, standards bodies, industry, civil society and technical experts should develop an Australian profile compatible with relevant international standards, including ISO/IEC 42001 for AI management systems, while reflecting Australian law, public- interest expectations and sectoral requirements.

  11. The profile should provide SME and low-risk baselines, enhanced controls for significant interests or scale, stronger expectations for agentic, public-sector, critical-infrastructure and security-sensitive systems, and reusable procurement, supplier, audit and regulatory evidence templates.

    The profile should provide a simple baseline for SMEs and low-risk adoption; enhanced controls for systems affecting significant interests or operating at scale; stronger expectations for agentic, public-sector, critical-infrastructure and security-sensitive systems; and reusable evidence templates for procurement, suppliers, internal audit and regulatory engagement.

  12. For sensitive, regulated or strategically significant deployments, have policy encourage patterns that preserve customer control, support data-residency and isolation requirements, minimise unnecessary data transit, retain auditable logs and make material supplier dependencies transparent.

    For sensitive, regulated or strategically significant deployments, policy should encourage patterns that preserve customer control, support data-residency and isolation requirements, minimise unnecessary data transit, retain auditable logs and make material supplier dependencies transparent.

  13. Government should demonstrate clear public purposes, human responsibility, accessible challenge and redress, documented risk decisions, independent review where warranted, and intelligible public reporting that protects security-sensitive information.

    Government should demonstrate clearly stated public purposes, meaningful human responsibility, accessible pathways for challenge and redress, documented risk decisions, independent review where warranted, and public reporting that is intelligible without exposing security-sensitive information.

Read the original

Open in the interactive guide