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 11

OpenAI

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.

Describes OpenAI's view that Australia can convert widespread AI adoption into productivity and skills gains. Covers four priorities: productivity and skills, consumer and youth safety, cybersecurity and national security, and infrastructure and copyright, including support for a balanced copyright framework and international AI Safety Institute cooperation.

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. Says skills policy should accompany adoption, with AI taught as an applied skill within a worker's chosen field.

    Skills policies therefore need to accompany adoption. AI should increasingly be taught as an applied skill within a worker’s chosen field, with training developed through partnerships among employers, workers, universities, technical and trade education providers and technology companies.

  2. Says useful AI capabilities need not be confined to large companies, high-income workers or major metropolitan areas.

    This matters in a country as geographically and economically diverse as Australia. Useful AI capabilities need not be confined to large companies, high-income workers or major metropolitan areas.

  3. Says broad access to AI must be paired with safeguards for vulnerable users, particularly young people.

    Broad access must also be safe access for vulnerable populations, particularly for young people.

  4. States that cyber security is an immediate priority for safety, national security and sovereignty, and that AI offers a temporary window for defenders.

    Cyber security is an immediate priority for safety, national security, and sovereignty. We are currently in a narrow defender’s window: a temporary opportunity to use AI to find and fix l ongstanding cyber weaknesses before attackers can exploit them at greater speed and scale.

  5. Argues that many of the most consequential AI safety risks cross borders, making international coordination increasingly important.

    There is also a broader question of AI safety as models become more capable. Many of the most consequential risks will not stop at national borders, making international coordination i ncreasingly important.

  6. Proposes a copyright framework allowing learning from publicly available information, rightsholder collaboration and protections against pirated content, infringing outputs and unlawful uses.

    OpenAI believes in a balanced copyright framework that allows models to learn from publicly available information, provides opportunities for collaboration with rightsholders, and maintains protections against pirated content, infringing outputs, and other unlawful uses.

Recommendations

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

  1. Teach AI within workers’ chosen fields, with training partnerships among employers, workers, universities, technical and trade education providers and technology companies.

    Skills policies therefore need to accompany adoption. AI should increasingly be taught as an applied skill within a worker’s chosen field, with training developed through partnerships among employers, workers, universities, technical and trade education providers and technology companies.

  2. Tailor AI regulation to its distinctive risks and beneficial uses in education, creativity, accessibility and skills development.

    AI is not social media, and regulations should account for its distinctive risks as well as beneficial uses in education, creativity, accessibility, and skills development.

  3. Make youth AI policies combine strong default safeguards with transparency, continuing evaluation and accountability, while preserving young people's ability to learn.

    Effective youth policies should combine strong default safeguards with transparency, continuing evaluation and accountability, while preserving young people’s ability to learn and prepare for the future.

  4. Require national security uses of AI to operate within Australian law, keep meaningful human judgement over consequential decisions and be subject to democratic oversight by institutions with technical expertise and appropriate access.

    National security uses must operate within Australian law, preserve meaningful human judgement over consequential decisions, and be subject to effective democratic oversight. The institutions responsible for that oversight should have the technical expertise and appropriate access needed to scrutinise how AI contributes to government decisions.

  5. Reinforce cooperation with other AI Safety Institutes.

    As a key democracy, Australia should reinforce cooperation with other AI Safety Institutes.

  6. Establish secure channels for sharing findings on advanced capabilities, vulnerabilities and safeguards, enabling Australian authorities to learn from overseas work and contribute expertise.

    Australia’s memorandum of understanding on AI safety and security collaboration is a good example of this type of international coordination. Secure channels for sharing findings on advanced capabilities, vulnerabilities and safeguards would help Australian authorities learn from work undertaken elsewhere and contribute their own expertise.

  7. Support joint evaluations and coordinated responses when AI risks cross borders.

    Joint evaluations and coordinated responses would be particularly valuable when risks cross borders.

  8. Extend international cooperation to shared technical standards for measuring capabilities, assessing risks and evaluating safeguards.

    This cooperation should also extend to shared technical standards. Common approaches to measuring capabilities, assessing risks and evaluating safeguards would make evidence easier to compare across countries and reduce duplicated work.

  9. Develop, with other countries, shared definitions, severity levels and reporting thresholds for significant AI incidents.

    Shared definitions, severity levels and reporting thresholds for significant AI incidents would also help countries respond together.

  10. Resource public agencies and critical infrastructure operators to prepare for AI-enabled attacks and rehearse responses through tabletop exercises.

    Australia should ensure that public agencies and critical infrastructure operators have the resources to prepare for AI-enabled attacks and rehearse their responses through tabletop exercises.

  11. Support public agencies and critical infrastructure defenders to prioritise internet-facing services, authentication, deployment and sensitive-information systems, then address vulnerability backlogs and verify fixes.

    Australia should ensure that public agencies and critical infrastructure operators have the resources to prepare for AI-enabled attacks and rehearse their responses through tabletop exercises. Defenders should assess internet-facing services, authentication, deployment systems and systems handling sensitive information first, then work through existing vulnerability backlogs and verify that fixes work.

  12. Facilitate AI infrastructure project development and construction as proposed in the National AI Plan.

    OpenAI supports the National AI Plan’s proposals to facilitate project development and construction.

  13. Adopt a balanced copyright framework that allows models to learn from publicly available information, with collaboration with rightsholders and protections against pirated content, infringing outputs and other unlawful uses.

    OpenAI believes in a balanced copyright framework that allows models to learn from publicly available information, provides opportunities for collaboration with rightsholders, and maintains protections against pirated content, infringing outputs, and other unlawful uses.

Read the original

Open in the interactive guide