Submission 273
ServiceNow
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 workflow platform provider argues that AI-cyber risk convergence is the most urgent AI risk and that government should adopt "governed autonomy" (agent identities, audit trails, human oversight, kill switches, an AI control tower, which is also how it describes its own platform). It recommends technology-neutral regulation on existing law, AI-layer data residency and skills investment.
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.
Argues that the convergence of AI capability with cyber risk is the most urgent AI risk, citing frontier models able to discover vulnerabilities and generate working exploits.
The most urgent AI risk facing Australia is not speculative. It is the convergence of AI capability with cyber risk, and it is happening now. Earlier this year the arrival of frontier models able to discover vulnerabilities and generate working exploits on their own marked a permanent shift
Recommends governed autonomy for Commonwealth adoption, where AI acts at machine speed while people define intent, supervise decisions and keep authority to intervene.
The operating model we recommend is governed autonomy: a system that senses its environment, decides with full business context, acts at machine speed, and governs every action, while people define intent, supervise classes of decisions, and keep the authority to intervene. Autonomous does not mean unattended.
Argues that every AI agent needs an auditable, least-privilege identity defined before it is switched on.
Every AI agent — like every person and every machine — needs an auditable identity and least-privilege access defined before it is switched on: what it can access, which privileges it holds, and how far its actions may reach.
Argues that data residency must hold at the AI layer, not only the hosting layer, and be treated as a first-order sovereignty question.
Data residency must hold at the AI layer. Commitments made at the hosting layer can be undermined if inference, model hosting or agent actions move data offshore. We recommend AI-layer data residency be treated as a first-order sovereignty question.
Says ASIC’s report found AI adoption outpaced governance updates and reminded organisations that obligations are technology-neutral and strong human oversight is expected.
ASIC’s Report 7985 found organisations are adopting AI faster than they are updating governance, and reminded them that existing obligations are technology-neutral and that strong human oversight is expected.
Argues that AI is more likely to augment than replace roles and that the binding constraint is skills rather than headcount.
AI is far more likely to augment roles than replace them; the binding constraint is skills, not headcount.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Treat AI and cyber risk convergence as a national-security priority, support the Office of AI's remit to coordinate frontier-model cyber risk, and define regulators' 'ability to respond' in operational terms (real-time detection, containment, reversibility).
Treat the convergence of AI and cyber risk as a national-security priority, support the Office of AI’s remit to coordinate frontier-model cyber risk across the economy, and define regulators’ “ability to respond” in operational terms — real-time detection, containment and reversibility (ToR l, m).
Make governed autonomy standard for Commonwealth AI: one control tower over every agent, identity-scoped least-privilege access, complete audit trails, human oversight and a kill switch on every agent.
Make governed autonomy the standard for Commonwealth AI use: a single AI control tower over every agent, identity-scoped least-privilege access, complete audit trails, human oversight, and a kill switch on every agent (ToR h).
Continuously monitor AI-agent identity and access limits and enforce them beneath the model; govern human, machine and AI-agent identities under one model.
Those limits must be monitored continuously and enforced beneath the model, so the system cannot do what it is not permitted to do, however it is prompted or manipulated. Human, machine and AI- agent identities should be governed under one model, not scattered across disconnected tools.
Enable agencies to stop or reverse agents instantly with enforceable off-switches and full reversibility when agents operate outside their intent.
Agencies must be able to stop or reverse an AI agent instantly; an enforceable off-switch and full reversibility, not a policy document describing what should happen, but a control that acts at machine speed the moment an agent operates outside its intent.
Keep AI regulation technology-neutral, risk-based and focused on deployment, on the basis that most existing law already applies, closing the operational-assurance gap rather than duplicating statute.
Keep AI regulation technology-neutral, risk-based and focused on deployment; most existing law already applies, so close the operational-assurance gap rather than duplicating statute (ToR i).
Treat AI-layer data residency as a sovereignty requirement and support domain-specific models on Australian data.
Treat data residency at the AI layer as a sovereignty requirement, and support domain-specific models trained on Australian data (ToR e, f).
Invest in skills tied to real demand and keep people at the centre of adoption.
Invest in skills tied to real demand and keep people at the centre, so the productivity benefit — and better service to Australians — is shared (ToR a, g).