Extend the AI Safety Institute's remit to harms from deployed AI
12 submissions. Each submission counts once for this shared ask. Counts do not measure importance or merit.
A shared ask groups requests for substantially the same action from three or more submissions when the group was created. Groups remain listed if withdrawals reduce their counts. AI models summarised recommendations, grouped them and checked the groups. The label summarises the requests; it is not a quote. A shared ask does not mean agreement on anything else.
Theme: Australian AI Safety Institute
Recommendations from submissions
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. Extracts are chosen by an AI model. Submissions are listed in submission number order; recommendations follow their order on each submission’s page.
ADEIA, UNSW and the ARC Centre for Next-Gen Architectural Manufacturing Submission 53
Have AI Safety Institute testing cover advanced/agentic design, asset-management and critical-infrastructure systems: leakage, prompt/data manipulation, unsafe tools, updates, audit-evidence loss and failure in degraded or disconnected conditions.
Australia's AI Safety Institute should develop formal sector partnerships with ASD, infrastructure agencies, regulators, universities and professional bodies. Testing should cover advanced and agentic systems used in design, asset management and critical infrastructure, including data leakage, prompt or data manipulation, unsafe tool use, model updates, loss of audit evidence and failure under degraded or disconnected conditions.
A program corrected obvious text-recognition errors in this quote for display.
Alannah & Madeline Foundation Submission 55
Attached paper: Explicitly include children’s safety, privacy and participation in the AI Safety Institute’s foundational research and regulator-engagement mandate.
The establishment of an AI Safety Institute. The Institute presents an opportunity to embed children’s rights into its foundational work programme. Its mandate to generate technical research into AI risks and harms, and to engage with regulators, should explicitly include children’s safety, privacy and participation as core priorities from the outset.
Institute for Integrated Economic Research - Australia Submission 65
Extend the AI Safety Institute's remit to population-level and institutional cognitive effects, including the Commonwealth's own adoption programme, and have the Committee examine whether its placement affords sufficient independence.
That the Australian AI Safety Institute’s remit be extended to cover population-level and institutional cognitive effects, including the Commonwealth’s own adoption programme, and that the Committee examine whether the Institute’s placement within the portfolio responsible for AI adoption and investment attraction affords it sufficient independence.
ValiDATA AI Submission 83
Ask the AI Safety Institute, with ASD and regulators, to test purchasing, payment and agent-to-agent interactions, and publish methods and findings where appropriate, including the limits of tested safeguards.
Recommend that the Government ask the AI Safety Institute, working with ASD and relevant regulators, to build on existing research through tests of purchasing, payment and interactions between agents. Publish methods and findings where appropriate, including the limits of tested safeguards.
Robotics Australia Group Ltd Submission 119
Include autonomous-system testing, critical supply-chain dependency identification and critical-infrastructure operators’ ability to maintain, update and audit deployed systems in the AI Safety Institute workstream.
Task the Australian AI Safety Institute with a dedicated Physical AI and Autonomous Systems workstream, working with Standards Australia, work health and safety regulators and critical-infrastructure agencies to develop practical assurance, cyber security and safety frameworks for AI-enabled machines operating in the physical world. This should include the ability to test autonomous systems, identify critical supply-chain dependencies and ensure operators of critical infrastructure retain the ability to maintain, update and audit deployed systems.
MDR Security Submission 137
Integrate AI safety and security into cyber security frameworks; include deployment risks in the AI Safety Institute’s remit, recognising frontier models may not suit every application.
Alsafety and security should be integrated into existing cyber security frameworks and methods rather than treated as a separate discipline, and the AI Safety Institute’s remit should cover the risks of deploying AI, not only the capabilities of frontier models — especially noting that frontier models may not be the solution for every potential AI application.
A program corrected obvious text-recognition errors in this quote for display.
Infosys Submission 141
Support the Australian AI Safety Institute to evaluate AI systems in deployment environments.
The Australian AI Safety Institute should be supported to evaluate AI systems in deployment environments.
Australian United Solidarity for Iran (AUSIRAN) Submission 279
Have the AI Safety Institute evaluate cumulative end-to-end misuse through multi-step scenarios involving state-linked surveillance, identity resolution, large-scale political profiling, coordinated influence, impersonation and tools identifying dissidents or diaspora members.
The Australian AI Safety Institute should evaluate end-to-end misuse, not only individual prompts. Safety evaluations should include multi-step scenarios involving state-linked surveillance, identity resolution, large-scale political profiling, coordinated influence operations, impersonation and tools designed to identify dissidents or diaspora members. Testing should assess the cumulative capability that can be assembled across a sequence of interactions.
Mr Simon Delaile Submission 367
Give the AI Safety Institute capacity to investigate interaction-level and real-world harms, including gradually emerging harms.
ensure the Australian AI Safety Institute has capacity to investigate interaction-level and real-world harms, including those that emerge gradually;
Ms Julie Green Submission 487
Fund the AI Safety Institute to build a unit with child protection and social work practice experience, able to independently evaluate and certify AI tools before use in child protection, family violence and similar services.
Fund the Australian AI Safety Institute to build a dedicated unit, staffed by people with child protection and social work practice experience rather than only AI researchers, with the authority to independently evaluate and certify AI tools before they can be used in child protection, family violence, and other services touching vulnerable Australians.
Mr Stephen Elliott Submission 508
Extend the AI Safety Institute's mandate to multi-agent, market-scale and social-media-scale failure modes, with resources.
AISI's mandate be extended to cover the evaluation of multi-agent, market-scale, and social-media-scale failure modes of AI, and resources be allocated to support this work;
Black Dog Institute Submission 549
Fund and empower the AI Safety Institute to protect individuals, including mental-health users: set safety standards, investigate harms, require incident reporting and hold AI providers accountable for risks to vulnerable users.
The Australian AI Safety Institute should have responsibility for protecting individuals from AI-related harms, including those arising from the use of AI for mental health support. This expanded remit should be supported by dedicated funding and powers to set safety standards, investigate harms, require incident reporting, and hold AI providers accountable for risks to vulnerable users.