Submission 528
Amplivy
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.
Amplivy proposes a technology-neutral standard for consequential Commonwealth and designated-sector actions, requiring verified human authorisation under institutions’ rules, including permitted prior authorisation. It describes fraud and agent risks and attaches a Treasury proposal for event-time presence verification. It discloses patented infrastructure not independently evaluated in deployment, and says the standard does not address criminal intent or deceived or coerced authorisers.
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.
Says identity, presence and authorisation are routinely conflated, and defines identity and presence as distinct things.
Three things are routinely conflated. Identity establishes who a person is. Presence establishes that the actual person is there, now, rather than someone or something using their credentials, face or voice.
Argues that sign-in authentication can confirm presence at login but does not by itself establish that the entitled person authorised a specific consequential action.
Authentication at sign-in can confirm that a person was present when they logged in. It does not, by itself, establish that the person entitled to authorise a specific consequential action has done so, least of all when an agent acts later on that person’s behalf, or with no person behind it at all.
Says government awareness of the reported incident depended on the developer's notification, citing Alastair MacGibbon, former head of the Australian Cyber Security Centre.
First, the government’s awareness depended on the developer’s notification: as Alastair MacGibbon, former head of the Australian Cyber Security Centre, observed, the victim did not know until the lab came and told it.
Argues that draft verification provisions should specify evidence that the entitled person approved a consequential action at execution.
None requires evidence that the person entitled to authorise a specific consequential action approved it at the moment it executed. The draft provisions still to be completed are the natural place to specify that outcome.
Says its standard does not address criminal intent or authorisation by deceived or coerced real people; institutions and law enforcement retain those controls.
It does not address a real person knowingly acting with criminal intent, or a legitimate person who is deceived or coerced into authorising an action themselves; institutions and law enforcement retain those controls.
Cites Anthropic's November 2025 disclosure of a state-sponsored, largely autonomous AI-enabled espionage campaign as an example of hostile exploitation.
In November 2025, Anthropic disclosed that a Chinese state-sponsored group had used its model to run a largely autonomous espionage campaign against around thirty targets, including government agencies and financial institutions, with the AI performing 80 to 90 per cent of the work.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Adopt an outcome standard: defined consequential actions on Commonwealth systems and in sectors already designated under the Scams Prevention Framework proceed only when verified identity and verified human presence show the entitled person authorised them, in accordance with the relying institution's own authorisation rules, whether requested by a person or an AI agent.
This submission asks the Committee to recommend a single, technology-neutral outcome standard: defined categories of consequential action against Commonwealth systems, and in the sectors already designated under the Scams Prevention Framework, should be gated before they execute, and should proceed only when verified identity and verified human presence, bound together, establish that the person entitled to authorise the action has done so, in accordance with the relying institution’s own authorisation rules, whether the request comes from a person or from an AI agent.
Have the Scams Prevention Framework codes and the foreshadowed mandatory Australian Standards for AI expressly address actions by AI agents and specify the verification outcome rather than any method or provider.
Recommend that the Scams Prevention Framework codes, including the verification provisions still to be completed, and the mandatory Australian Standards for AI the government has foreshadowed, expressly address actions taken by AI agents and specify the verification outcome rather than any method or provider.
Have agencies apply the standard to their own consequential transactions, including myGov account linking and payment changes, and to actions by AI agents they deploy.
Recommend that agencies apply it to their own consequential transactions, including myGov account linking and payment changes, and to actions taken by AI agents they deploy.
Run an independently evaluated Commonwealth pilot with published acceptance criteria, including resistance to bypass and replay, compromised devices, latency, accessibility, service continuity and privacy.
Recommend an independently evaluated Commonwealth pilot with published acceptance criteria, including resistance to bypass and replay, handling of compromised devices, latency, accessibility, service continuity and privacy.
Put independent controls at the moment of consequential actions, alongside mandatory, time-bound incident reporting by AI developers.
Recommend independent controls at the moment of consequential actions, alongside mandatory, time-bound incident reporting by AI developers.
Require any standard to be met without retaining identity documents or biometric records, with retained proof and its keys held in Australia under the relying institution's control.
Recommend that any standard be met without retaining identity documents or biometric records, and that retained proof, and the keys that secure it, be held in Australia under the relying institution’s control.
Have operators of critical infrastructure and systems of national significance apply the standard to privileged and consequential actions, independent of any single technology vendor.
Recommend that operators of critical infrastructure and systems of national significance apply the standard to privileged and consequential actions, independent of any single technology vendor.
Apply the standard first to defined actions, for example payment-destination changes or above-threshold payments, account/device linking, privileged access and protected-information release, including such actions requested by AI agents.
The standard should apply first to a defined set of consequential actions, for example: • changing the destination of a payment, or approving a payment above a set threshold; • linking a new account or device to an existing identity, such as a myGov link; • granting privileged access to a system, or releasing protected information; • any such action requested by an AI agent rather than directly by a person.
Attached Treasury proposal: require event-time human presence and voluntary control alongside onboarding checks; keep controls technology-neutral, deterministic where practicable, complementary to AML/CTF diligence, data-minimising, compatible with privacy and encryption without content inspection, and auditable for disputes.
Where a sector code requires verification, a regulated entity should be able to establish, at the time a protected event executes, that a real human is present and in voluntary control, and should not rely solely on identity checks performed at onboarding. Controls implementing this principle should: • be technology-neutral, specifying the outcome rather than any particular method or provider; • establish presence deterministically where practicable, rather than inferring it; • complement, not replace, existing customer due diligence under AML/CTF law; • minimise the collection and retention of identity data; • operate consistently with privacy law and with encryption, without inspecting message content; • produce an auditable record capable of supporting internal and external dispute resolution.