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 545

Salesforce

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.

Salesforce, a cloud and AI software provider, advocates tailored, risk-based AI regulation that distinguishes developers from deployers. It addresses SME adoption, public sector AI use with human oversight, AI Safety Institute interoperability, a nuanced view of sovereign capability and hosting a global AI summit; some asks name Salesforce's own products or programs.

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. States that it supports tailored, risk-based AI regulation that differentiates contexts and uses.

    Salesforce supports tailored risk-based AI regulation that differentiates contexts and uses of the technology and ensures the protection of individuals, building of trust, and innovation.

  2. Argues that a one-size-fits-all regulatory approach may hinder innovation and delay adoption.

    A one-size-fits-all regulatory approach may hinder innovation, disrupt healthy competition, and delay the adoption of technology that consumers and businesses around the world are already using to boost productivity.

  3. Argues that regulation should assign different guardrails to AI developers and deployers.

    Regulation should distinguish between developers and deployers and assign guardrails accordingly. Generative AI developers should be accountable for how their models are trained and the data used to train them; those deploying the technology control how it's used and should be held to rules that reflect that specific use.

  4. Argues that government AI should not make decisions with legal or similarly significant effects on citizens unless a human makes the final decision.

    AI should not be used to automatically make decisions with legal or similarly significant effects on citizens (such as denying benefits or imposing penalties) unless a human being makes the final decision.

  5. Argues that strategic sovereignty means control over how digital assets and algorithms operate within Australian jurisdiction, not isolated national systems.

    Strategic sovereignty does not mean building isolated national systems; it means retaining the authority, governance, and technical architecture required to control how digital assets and algorithms operate within Australian jurisdiction.

  6. Says research finds SMEs using AI report expectations of business change, easier scaling and improved margins.

    Research reveals that SMEs with AI have a bright outlook: 78 percent say it will be a game-changer for their company, 87 percent say AI helps them scale operations, while 86 percent see improved margins.

Recommendations

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

  1. Consider supporting SME AI adoption nationally through the National AI Centre in partnership with industry, providing practical tips and examples from other small businesses.

    Recommendation 1: The Australian government consider supporting SME and AI nationally through Australia's National AI Centre in partnership with industry to provide practical tips for SMEs to use AI to improve their businesses and practical examples from other small businesses.

  2. Provide guidance to SMEs to prioritise AI governance, explainability and compliance with the Privacy Act to protect sensitive information.

    Recommendation 2: Guidance be provided to SMEs to prioritise AI governance, explainability, and compliance with the Australian Privacy Act to protect sensitive information.

  3. Maintain meaningful human participation in government AI, augmenting public servants' judgement, particularly for human rights and trust; require a human final decision for legal or similarly significant effects on citizens.

    The use of AI in government should maintain meaningful human participation, which serves to augment the judgment of public servants rather than replacing it, particularly for decisions that may compromise human rights and public trust. Areas where AI use should be avoided: • Fully Autonomous Decision-Making. AI should not be used to automatically make decisions with legal or similarly significant effects on citizens (such as denying benefits or imposing penalties) unless a human being makes the final decision.

  4. Avoid government AI explicitly predicting or categorising individuals by protected characteristics such as race, gender, disability or religious beliefs.

    Discriminatory Profiling. AI should not explicitly predict or categorize individuals based on protected characteristics like race, gender, disability, or religious beliefs.

  5. Shift focus to citizen outcome-driven assurance, reframing AI assurance as a mechanism to safely unlock public value, boost productivity and improve frontline service delivery.

    Recommendation 3: The Australian government shift focus to citizen outcome-driven assurance, reframe AI assurance as a mechanism to safely unlock public value, boost productivity, and improve frontline service delivery.

  6. Have regulation focus most on high-risk applications, especially those that could cause significant harm or affect someone's rights and freedoms.

    Risk-based AI regulation should focus most on high-risk applications — especially those that could cause significant harm or affect someone's rights and freedoms.

  7. Distinguish developers and deployers: hold generative-AI developers accountable for model training and training data, and apply deployment rules reflecting each specific use.

    Regulation should distinguish between developers and deployers and assign guardrails accordingly. Generative AI developers should be accountable for how their models are trained and the data used to train them; those deploying the technology control how it's used and should be held to rules that reflect that specific use.

  8. Ensure generative-AI rules address personal-data use and protection during model training and throughout the AI ecosystem.

    Rules specific to generative AI should address how personal data is used and protected when training future models, and how it's safeguarded across the AI ecosystem.

  9. Require explanations of AI-driven recommendations and disclosure of whether people are interacting with humans or simulated personas.

    People should understand the "why" behind AI-driven recommendations, and should know when they're interacting with a human versus a simulated persona.

  10. Use AI impact assessments for high-risk-system accountability; any licensing or notifications should be risk-based, balancing oversight with market entry, competition and innovation.

    AI impact assessments help build accountability and trust for high-risk systems. Licenses or notifications can support compliance, but should follow a risk-based approach that balances oversight with market entry, competition, and innovation. Risk-management frameworks like NIST's offer a solid foundation for the field.

  11. As regulators develop new guidance, first assess whether existing law already addresses the concern before layering on something new.

    As regulators develop new guidance, they should first assess whether existing law already addresses the concern before layering on something new.

  12. Make AI regulation globally interoperable, durable and flexible, focused on ethical development and deployment rather than a particular technology at a particular moment.

    Given the pace of innovation, AI regulation should be globally interoperable, durable, and flexible — focused on the ethical development and deployment of AI systems, not on a specific technology at a specific moment in time.

  13. Consider Singapore's AI Verify, an AI governance testing framework of which Salesforce is a founding member, as a model for Australia.

    Other models for Australia to consider is AI Verify, a subsidiary of the Infocommunications Media Development Authority of Singapore (IMDA). It is an AI governance testing framework and software toolkit that validates the performance of AI systems against a set of internationally recognised principles through standardised tests and is consistent with international AI governance frameworks. Salesforce is one of the founding members.

  14. Have the AI Safety Institute prioritise standardised, empirical evaluations for autonomous AI systems, such as continuous red-teaming, threat modelling for model exploitation and adversarial prompt vulnerability assessments.

    AAISI should prioritise standardised, empirical evaluations for autonomous AI systems (e.g., continuous red-teaming, threat modelling for model exploitation, and adversarial prompt vulnerability assessments).

  15. Have the AI Safety Institute maintain international interoperability to prevent regulatory fragmentation while safeguarding national sovereign interest.

    Recommendation 4: AAISI maintains international interoperability to prevent regulatory fragmentation while safeguarding national sovereign interest.

  16. Mandate in national security frameworks that agentic deployments have embedded governance layers (citing Salesforce's own Einstein Trust Layer) enforcing strict access controls, zero data retention by underlying LLMs and real-time toxicity and security guardrails.

    National security frameworks must mandate that agentic deployments feature embedded governance layers—such as Salesforce’s Einstein Trust Layer— that enforce strict access controls, zero data retention by underlying LLMs, and real-time toxicity and security guardrails.

  17. Have regulators ensure sovereign, domain-specific AI models operate under Zero Copy Architecture, querying local datasets without exposing sensitive national security data to external or foreign training pipelines.

    Regulators must ensure that sovereign, domain-specific AI models operate under Zero Copy Architecture, querying local datasets securely without exposing sensitive national security data to external or foreign training pipelines.

  18. Have the AI Safety Institute and national security regulators establish direct, continuous technical exchanges with major enterprise AI providers for early visibility into capability jumps, safety testing benchmarks and agentic operational models.

    Recommendation 5: The AAISI and national security regulators establish direct, continuous technical exchanges with major enterprise AI providers to gain early visibility into capability jumps, automated safety testing benchmarks, and agentic operational models.

  19. Use trusted security alliances, including Five Eyes, Quad partners and bilateral technology agreements, to support supply-chain resilience, shared threat intelligence and interoperable regulatory standards.

    Australia should leverage its trusted international security alliances—such as the Five Eyes network, Quad partners, and bilateral technology agreements—to ensure supply chain resilience, shared threat intelligence, and interoperable regulatory standards.

  20. Define sovereign AI settings around security, control and operational resilience rather than absolute technological isolation, recognising international collaboration with partner nations and trusted providers.

    Recommendation 6: Adopt a nuanced definition of sovereign capability. The Australian Government define sovereign AI settings around security, control, and operational resilience rather than absolute technological isolation—recognising that international collaboration with like-minded partner nations and trusted technology providers is critical to national security and innovation.

  21. Have regulators establish clear, standardised data classification tiers across Commonwealth and State agencies, using a modern risk-based approach.

    Recommendation 7: Implement Modern Risk-Based Data Classification. Regulators establish clear, standardised data classification tiers across Commonwealth and State agencies.

  22. Be a future host of the AI Global Summit bringing together government and technology leaders.

    Recommendation 8: The Australian government be a future host of the AI Global Summit which brings together government and technology leaders from around the globe

Read the original

Open in the interactive guide