Submission 135
Mimecast
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.
Mimecast, a cybersecurity company, argues that generative AI has outpaced human and technical detection of fraud and impersonation, and that frontier AI assessment is lagging. It recommends targeted liability amendments, a stronger AISI, provenance standards, critical infrastructure supply chain rules, and systemic safeguards over reliance on individual vigilance.
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.
States its opinion that generative AI has outpaced detection methods, making AI-enabled attacks the most urgent area for action.
In our opinion, generative AI has compressed the cost, speed and personalisation of attacks to a point that outpaces the detection methods, both human and technical, that organisations have relied on for the past decade. This area represents the most urgent area for action.
Argues that AI supply chain dependencies are poorly mapped in most organisations' supplier risk registers.
the AI supply chain itself such as the training data, pre-trained components, third-party APIs and agentic tool integrations, introduces dependencies that are poorly mapped in most organisations' existing supplier risk registers, and critical infrastructure operators are already grappling with increasingly complex, cross-border, cyber-physical supply chains.
States that frontier AI models are the biggest concern for cybersecurity leaders and that assessment capability is not keeping pace.
Frontier AI models represent the biggest risk/concern for cybersecurity leaders across Australia. Additionally, the assessment capability is not keeping pace with model capabilities.
Argues that AISI's advisory rather than regulatory mandate limits its ability to turn identified risks into enforceable safeguards.
second gap is associated with AISI's current mandate which is advisory rather than regulatory. This limits its ability to translate identified risks into enforceable safeguards for high-risk, agentic use cases.
Argues that the absence of provenance standards both unsettles copyright and enables brand impersonation.
From a cybersecurity standpoint, the same absence of provenance standards that unsettles copyright also enables bad actors to reproduce an organisation's brand, voice and likeness convincingly in phishing, invoice fraud and impersonation campaigns.
Argues that resilience depends on systemic safeguards that reduce reliance on individual judgement, with training as a supporting layer.
Resilience instead depends on reducing how much any single decision relies on individual judgement, through friction and verification built into high-risk workflows, clear identity and access controls covering both human and AI activity, and technology controls that act ahead of the point of human decision, with awareness and training operating as a supporting layer rather than the primary safeguard.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Clarify liability and attribution for harms from autonomous or agentic AI decision-making, through targeted amendments rather than a standalone AI Act.
Clarify liability and attribution for harms arising from autonomous or agentic AI decision-making, through targeted amendments rather than a standalone AI Act, consistent with the Government's technology-neutral stance.
Give AISI a formal, resourced evaluation and assurance function for agentic and frontier systems used in security-sensitive contexts, with a clear escalation path to sector regulators.
Give AISI a formal, resourced evaluation and assurance function for agentic and frontier systems used in security-sensitive contexts, with a clear escalation path to sector regulators.
Commission a standing, cross-portfolio gap analysis, updated annually, of where privacy, consumer, telecommunications and online safety law fail to address generative AI-enabled fraud, impersonation and synthetic content.
Commission a standing, cross-portfolio gap analysis (updated annually) of where privacy, consumer, telecommunications and online safety law fail to address generative AI-enabled fraud, impersonation and synthetic content.
Support development of content provenance and authenticity standards (e.g. C2PA-aligned metadata) for copyright attribution and brand-impersonation risk.
Support development of content provenance and authenticity standards (e.g. content credentials/C2PA-aligned metadata) as a shared solution to both copyright attribution and brand- impersonation risk.
Extend any licensing or transparency obligations on frontier model developers for training data to also cover disclosure of safeguards against misuse of a person's or organisation's likeness, voice and brand assets.
Ensure any licensing or transparency obligations placed on frontier model developers for training data also extend to disclosure of safeguards against the misuse of a person's or organisation's likeness, voice, and brand assets.
Legislate a criminal offence for knowingly or recklessly using copyrighted, personal or brand-related material to train AI models to cause harm, with a civil penalty regime for negligent conduct below the criminal threshold.
Legislate a criminal offence for knowingly or recklessly using copyrighted, personal, or brand-related material to train AI models for the purpose of causing harm, paired with a civil penalty regime for negligent conduct that falls below the criminal fault threshold.
Fund a national, standing capability (via the ACSC/AISI) to test and publish real-world detection accuracy of deepfake and synthetic-voice tools against current-generation models.
Fund a national, standing capability (via the ACSC/AISI) to test and publish the real-world detection accuracy of deepfake and synthetic-voice tools against current-generation models.
Mandate or strongly incentivise multi-factor verification for high-value payment and credential-reset workflows, including a delay in payment time to enable cancellation.
Mandate or strongly incentivise multi-factor verification for high-value payment and credential-reset workflows, including a delay in the payment/transaction time to enable cancelation.
Extend the Online Safety Act's rapid-takedown and classification mechanisms to AI-generated deepfakes targeting children and vulnerable Australians, with a dedicated reporting pathway distinct from scam reporting.
Extend the Online Safety Act's rapid-takedown and classification mechanisms explicitly to AI-generated deepfakes targeting children and vulnerable Australians, with a dedicated reporting pathway distinct from general scam reporting.
Support continued, well-funded public awareness campaigns on AI-enabled scams, tailored for small business, older Australians and young people.
Support continued, well-funded public awareness campaigns (building on past ACSC and Federal Government efforts) specifically addressing AI-enabled scams, tailored for small business, older Australians and young people.
Extend the public awareness campaign to target care providers of the elderly and NDIS participants on scams and on behaviours suggesting the person in their care may have been targeted.
Extend the public awareness campaign to target care providers to the elderly and NDIS participants regarding scams and to understand behaviours that may suggest that the individual under their care may have been targeted.
Review the National Curriculum to include education on risk and harm arising from AI, possibly within the General Capability on digital literacy.
Review the National Curriculum to include education around risk and harm arising from AI. This could be included in the General Capability - digital literacy.
Extend the Security of Critical Infrastructure Act’s risk management requirements to AI supply chain dependencies: training-data and model provenance, third-party inference and agentic tool access.
Extend the Security of Critical Infrastructure Act's risk management program requirements to explicitly cover AI supply chain dependencies (training data provenance, model provenance, third-party inference and agentic tool access).
Establish sector-specific guidance, coordinated through the ACSC, on mapping and attesting AI supply chain dependencies, building on existing frameworks rather than a parallel scheme.
Establish sector-specific guidance, coordinated through the ACSC, on mapping and attesting AI supply chain dependencies, building on existing frameworks rather than creating a parallel scheme.
Expand IRAP to include critical infrastructure entities, to ensure a baseline and that supply chain risks are assessed.
Expand IRAP to include critical infrastructure entities to ensure a baseline and that supply chain risks are assessed.
Direct investment toward sovereign, AI-assisted threat detection capability for government and critical infrastructure operators.
Direct investment toward sovereign, AI-assisted threat detection capability for government and critical infrastructure operators, recognising that AI- enabled attacks now frequently outpace human-speed response.
Formalise information-sharing between the ACSC, AISI and the intelligence community on AI-enabled attack techniques observed against Australian targets, so lessons reach defenders faster than through general threat bulletins.
Formalise information-sharing between the ACSC, AISI and the intelligence community specifically on AI- enabled attack techniques observed against Australian targets, so lessons reach defenders faster than they currently do through general threat bulletins (information sharing has matured in recent times).
Resource AISI to adopt, and contribute Australian data to, internationally recognised frontier-risk evaluation frameworks covering cyber-offense uplift and autonomous capability, rather than developing a bespoke Australian methodology in isolation.
Resource AISI to adopt (and contribute Australian data to) internationally recognised frontier-risk evaluation frameworks covering cyber-offense uplift and autonomous capability, rather than developing a bespoke Australian methodology in isolation.
Establish a standing technical liaison between AISI, the Australian Signals Directorate and the National Intelligence Community, to translate frontier evaluation findings into practical guidance for regulators and critical infrastructure operators.
Establish a standing technical liaison between AISI, the Australian Signals Directorate and the National Intelligence Community, with a mandate to translate frontier evaluation findings into practical guidance for regulators and critical infrastructure operators.
Review at least twice a year whether the pace of frontier model capability has outstripped the assumptions underpinning Australia's technology-neutral regulatory posture.
Review at least twice a year, whether the pace of frontier model capability has outstripped the assumptions underpinning Australia's current technology-neutral regulatory posture, given how quickly this evidence base is moving.
Create a funded AI research mechanism linking Australian tertiary institutions, coordinating resources to focus effort, avoid duplication and move Australia to the forefront of the technology.
Create a funded research mechanism that is focused on AI research. The aim would be to establish a strong collaboration network across Australian tertiary institutions. The aim would be to move Australia to the forefront of the technology and co-ordinate resources to ensure focus and no duplication of effort.
Include AI-security literacy as a named component of any Commonwealth AI workforce or skills initiative.
Include AI-security literacy as a named component of any Commonwealth AI workforce or skills initiative, not solely a data-science or engineering capability.
Treat systemic safeguards (verification friction, identity and access controls for humans and AI agents, workflow design reducing reliance on vigilance) as at least equal in weight to awareness, training and regulatory controls.
Ensure the Committee's final recommendations treat systemic safeguards, including verification friction, identity and access controls covering both human and AI-agent activity, and workflow design that reduces reliance on individual vigilance, as being of at least equal weight to awareness, training and regulatory controls.