Submission 39
Australian Institute of Insider Threats
Submitter group: Research institutes.
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.
Argues that as agentic AI moves from advising to acting, organisations should govern the access, authority and autonomy delegated to AI. Proposes a five-element exposure model, seven governance questions and four considerations for the Committee, including identifiable human accountability.
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 governance should cover the access, authority and capability organisations delegate to AI, not only the technology itself.
As AI moves from advising organisations to acting on their behalf, governance needs to consider the access, authority and organisational capability being delegated, not simply the technology receiving it.
Argues that an AI system that recommends an action differs fundamentally from one with authority to act.
An AI system that recommends an action is fundamentally different from one that has the authority to act.
Argues that exposure depends on what the organisation allows the AI to access and do, not only on what the AI can do.
The level of exposure therefore depends not only on what the AI can do, but on what the organisation allows it to access and do.
Argues that individually authorised permissions may combine to create capability that was never intended.
individually authorised permissions may collectively create a level of capability that was never intended.
Argues that greater AI autonomy should not obscure who is responsible for consequential AI activity.
Greater AI autonomy should not obscure responsibility. An identifiable person or organisational function should remain accountable for consequential AI activity.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Ask the Committee to consider organisations documenting delegated AI authority, including what AI can access, decide, initiate, change or communicate.
Organisations should understand and document what authority has been delegated to AI, including what it can access, decide, initiate, change or communicate.
Asks the Committee to consider that organisations should understand the cumulative capability created by combined permissions.
Permissions that appear appropriate when considered individually may, when combined, create unintended capability. Organisations should understand what an AI system can do because of its combined access and authority.
Ask the Committee to consider increasing oversight, intervention and assurance as the potential consequences of AI actions increase.
The greater the potential consequence of an AI system's actions, the greater the level of oversight, intervention and assurance that may be required.
Ask the Committee to consider keeping an identifiable person or organisational function accountable for consequential AI activity.
An identifiable person or organisational function should remain accountable for consequential AI activity.