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 392

Ms Alana Kane

Submitter group: Individuals.

Position labels are not shown for individual submitters.

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.

Argues for mandatory frontier-AI testing, high-risk safeguards, incident reporting, funded safety oversight and meaningful human control before a serious incident. Supports AI investment and benefits while recommending proportionate, risk-based regulation, international cooperation, ongoing review and protection of children and future generations.

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 Australia should reconsider the balance between voluntary guidance and enforceable safeguards, noting the Government has said it will not proceed with its earlier mandatory guardrails proposals.

    The Government previously consulted on mandatory guardrails for high- risk AI but has subsequently stated that it will not proceed with those previous proposals at this time. (Industry, Science and Resources) I believe this is an area where Australia should reconsider the balance between voluntary guidance and enforceable safeguards.

  2. Recommends mandatory safety requirements for frontier AI, including rigorous pre-deployment testing for dangerous capabilities.

    I recommend that Australia establish mandatory safety requirements for highly capable or “frontier” AI systems. Before particularly capable systems are deployed at scale, developers should be required to undertake rigorous testing for dangerous capabilities and foreseeable failure modes.

  3. Recommends that regulators and AI safety institutions have sufficient funding, expertise and model access to assess risks independently.

    The Government should ensure that regulators and Australia’s AI safety institutions have sufficient funding, technical expertise and access to advanced models to independently assess emerging risks.

  4. Recommends a mandatory reporting framework for significant AI safety incidents and near misses.

    Australia should establish a mandatory reporting framework for significant AI safety incidents and near misses.

  5. Recommends that government policy distinguish nominal human oversight from meaningful human control.

    Government policy should distinguish between nominal human oversight and meaningful human control.

  6. Argues for light-touch regulation of low-risk AI, stronger requirements for higher-risk AI, and the highest testing, monitoring and oversight for highly capable frontier AI with potentially systemic consequences.

    A sensible approach would be: Low-risk AI → light-touch regulation Higher-risk AI → stronger requirements Highly capable frontier AI with potentially systemic consequences → the highest level of independent testing, monitoring and oversight

Recommendations

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

  1. Require rigorous developer testing before particularly capable systems deploy at scale, covering relevant cyber, autonomous, deception/manipulation, safeguard-circumvention, external-access, biological/chemical, self-replication/modification/improvement, emergent-capability and loss-of-control risks.

    Before particularly capable systems are deployed at scale, developers should be required to undertake rigorous testing for dangerous capabilities and foreseeable failure modes. This should include, where relevant: • cyber security capabilities; • autonomous operation and the ability to pursue goals without continuous human direction; • deception and manipulation; • ability to circumvent safeguards; • ability to acquire or maintain access to external systems; • biological or chemical misuse risks; • ability to replicate, modify or improve aspects of their own operation; • unexpected emergent capabilities; • and scenarios involving loss of meaningful human control.

  2. Require independent safety evaluation, with appropriate confidentiality protections for commercially sensitive information, rather than testing solely by developers.

    Testing should not be solely conducted by the companies developing the systems. There should be independent evaluation, with appropriate confidentiality protections for commercially sensitive information.

  3. Avoid unnecessary innovation restrictions for low-risk AI; require stronger safeguards as capabilities grow, particularly for autonomy, sensitive-system access, code execution, cyber operations, critical infrastructure and materially influential decisions.

    For ordinary, low-risk uses of AI, regulation should not unnecessarily restrict innovation. However, as AI systems become substantially more capable, particularly where they can operate autonomously, access sensitive systems, write and execute code, conduct cyber operations, interact with critical infrastructure or materially influence important decisions, stronger mandatory safeguards should apply.

  4. Ensure regulators and Australia's AI safety institutions have sufficient funding, technical expertise and advanced-model access to independently assess emerging risks.

    The Government should ensure that regulators and Australia’s AI safety institutions have sufficient funding, technical expertise and access to advanced models to independently assess emerging risks.

  5. Require companies to report significant AI incidents and near misses, including significant real-world harm, unanticipated behaviour, circumvented safety controls, unauthorised autonomous actions, cyber/privacy incidents and other systemic-risk information.

    Australia should establish a mandatory reporting framework for significant AI safety incidents and near misses. This could operate similarly to reporting systems used in aviation, cyber security and other safety-critical industries. Companies should be required to report serious incidents such as: • AI systems causing significant real-world harm; • systems behaving in ways that developers did not anticipate; • successful circumvention of important safety controls; • significant autonomous actions outside authorised parameters; • serious cyber or privacy incidents involving AI; • or other events that could provide important information about systemic AI risks.

  6. Distinguish nominal human oversight from meaningful human control in government policy.

    Government policy should distinguish between nominal human oversight and meaningful human control.

  7. For high-risk applications, require organisations to demonstrate accountability, permitted AI decisions, human-approval points, stopping and access-revocation mechanisms, decision auditing and post-deployment monitoring.

    For high-risk applications, organisations should be required to demonstrate: • who is accountable for the system; • what decisions the AI is permitted to make; • when human approval is required; • how the system can be stopped; • how access can be revoked; • how decisions can be audited; • and how the system is monitored after deployment.

  8. Work closely with international partners to share emerging-capability, incident and effective safety-practice information.

    Australia should also work closely with international partners to share information about emerging capabilities, incidents and effective safety practices.

  9. Ongoing review of regulation as AI capabilities develop.

    Ongoing review of regulation as AI capabilities develop.

  10. Explicit consideration of the interests and safety of children and future generations.

    Explicit consideration of the interests and safety of children and future generations.

  11. Consider whether existing regulatory frameworks protect children adequately from immediate AI harms and longer-term systemic risks.

    I encourage the Committee to consider whether existing regulatory frameworks adequately protect children from both immediate AI harms and longer-term systemic risks.

  12. Apply light-touch rules to low-risk AI, stronger requirements to higher-risk AI, and the highest independent testing, monitoring and oversight to highly capable frontier AI with potentially systemic consequences.

    A sensible approach would be: Low-risk AI → light-touch regulation Higher-risk AI → stronger requirements Highly capable frontier AI with potentially systemic consequences → the highest level of independent testing, monitoring and oversight

Read the original

Open in the interactive guide