Mandatory Guardrails for High Risk AI – Submission
Stirling & Rose urges the Government to prioritise robust public discussion on AI regulation and Lawful AI Critical Infrastructure which foreshadows the capabilities, market/societal implications and risks of AI as it will be in 10 – 20 years, rather than as it is now or in the near term. .
Executive Summary – Prepare Now for the AI Future
To access the full version of Stirling and Rose’s submission please click: Mandatory Guardrails for High-Risk AI
The future – AI in everything
Stirling & Rose urges the Government to prioritise robust public discussion on AI
regulation and Lawful AI Critical Infrastructure1 which foreshadows the capabilities,
market/societal implications and risks of AI as it will be in 10 – 20 years, rather than as it
is now or in the near term.
Recent discussions on how to regulate AI, including the discussions occurring as a result
of the current Consultation, are predominantly focused on the current state of AI. Just as
a good AFL player anticipates the future moves on the field and positions themselves
ahead of the play, we urge Australian policymakers and regulators to prepare for where
AI is heading, including moving to develop Lawful AI Critical Infrastructure.
The current AI tools and platforms are controllable, identifiable and alienable as an AI tool
or platform – for example ChatGPT is an AI platform. AI as an alienable “other” will not be
the case for long. In our thought leadership, advisory and policy work, our experience
strongly suggests that to “play the movie forward” is a future state where AI is
embedded in everything – including us, our institutions, our consumer goods – every
sector and every industry – from our fridge to our cars, our money and our contracts.
This might take five years, this might take 15 years, it is however, inevitable. In this
postulated future, regulation of AI as a technology is likely to apply to regulate everything
– every product, every service. We also postulate that Lawful AI Critical Infrastructure will
be key to respecting the rule of law in an AI-integrated future.
Good law is technology agnostic. Law “parametrised” by the current shape of a specific
technology, particularly a rapidly transforming technology such as AI, will be quickly
found to be not fit for purpose. Any AI regulation must take into account the coming
inability to alienate an AI tool from an underlying asset, entity, or even biological unit.
Responsible AI Problem
To cast forward, Australia should be preparing for a future where increasingly capable,
agentic AIs require diminishing human control and oversight over time and are ultimately
capable of autonomously operating enterprises which deliver economic value and in
which humans invest as shareholders. These AIs will be capable of autonomous
evolution, self-replication and creating or “digitally birthing” and nurturing new, more
specialised AIs optimised for specific tasks or goals.
This raises the responsible AI problem, that all existing legislation is predicated on
regulatory non-compliance being ultimately sheeted back to a responsible human
person. By comparison, AI is not responsive to the human-centric sanctions underpinning
traditional regulatory structures such as shame, disqualification, state-sanctioned
violence and imprisonment.
We postulate new measures will be required to combat the responsible AI problem, such
as the granting of legal personality to AIs through a new legal entity, the AI Organisation
(AIO). However, the granting of such legal status should be conditioned upon satisfaction
of certain requirements e.g. registration with appropriate agencies, unique identifier as
an AIO, evidence of de minimis functionality, effective sanctions for aberrant behaviour,
economic reserves, adjustments for intent.
Going forward, another mitigating tool to also help address the responsible AI problem is
the smart legal contract2. Smart legal contracts hosted on secure (likely critical)
infrastructure can act as an AIO governance tool with greater ability to moderate the
complexity of risk allocation, rules of engagement, obligations and remedies between
differing members/shareholders and regulatory bodies in respect of the AIO and its
individual assets.
The increasing capability and multifaceted nature of AI demands an adaptable, futurefocused
policy response and regulatory framework.
The future of AI in Australia hinges on our collective resolve to forge a path that is both
progressive and conscientious, reflecting a commitment to civil society, progress and
ethical stewardship in the age of intelligence machines.
Responsible Data underpins Responsible AI
Responsible and authenticated data is a precondition to responsible AI. Data is the
indispensable resource underpinning an AI-driven economy. Accurate, fair, lawful AI
needs vast amounts of high-quality data, with trusted provenance and clear legal
attribution. Without data, AI cannot be developed. Without fresh, current data, existing
AI applications fall into the error and obsolescence of model drift. Without contextually
relevant, representative data, AI struggles to generalise from training data to real-world
settings, resulting in errors of accuracy, failures of fairness and reinforcement of harmful
biases and discrimination. Consideration of the legal requirements and frameworks for
data is time-critical as data available for AI training becomes increasingly scarce.
Discussions on AI regulation predominantly focus upon the EU AI Act. However, we
consider that the European Data Act is an important, though frequently overlooked,
touchstone in advancing AI regulation. Regulating and standard setting for AI is a complex
and dynamic endeavour. By comparison, the understanding of data is mature, although
data value is not yet fully realised.
The recognition of data as a strategic asset should be a cornerstone of AI policy in
Australia. Lawful data collection, control, authentication and access are core
requirements in the regulatory shepherding of responsible AI outcomes. Regulatory
planning and development of infrastructure around public and private data rights is
foundational to ensuring that lawful collection and access to data is protected and
monopoly-style aggregations of data are kept in check.
Lawful AI Critical Infrastructure
Our Australian economy is increasingly digital and real-time. AI will accelerate this.
Solutions to manage the digital nature and speed of an economy where AI exists at mass
scale will be necessary for legal persons to continue to exercise legal rights and access
legal remedies. We consider that Australia, like all sovereign nations, will need to invest
in smart legal contract infrastructure to support the collection of lawful, authentic, quality
data and the assetisation of the data (such as contracts) as second-generation digital
assets.
This infrastructure will need to be developed with an eye to data as sovereign wealth and
the ability to govern and effectively sanction increasingly capable and autonomous AI
over the coming years.
We strongly recommend that Australia consider the creation and resourcing of a Lawful
AI Critical Infrastructure Advisory Committee that works closely with or is a committee
governed by the ACCC. We recommend Australia establish smart legal contract
infrastructure to support the authentication, digitisation and assetisation of lawful data
and legal agreements as second-generation digital assets.
Cautious and Prudent Domain Specific Approach
In the immediate term, Stirling & Rose recommends a cautious and prudent approach to
AI regulation focused on leveraging existing laws and structures and making necessary
changes to address concrete (rather than speculative) aspects of harm or negative
impact. We consider that Australia should observe, evaluate and participate in global
review of regulatory and similar initiatives (e.g. standard setting) implementation in other
jurisdictions to ascertain whether such approaches might be suitable for adoption in
Australia. At this juncture, we consider there will be substantive innovation, policy and
legal debt associated with Australia implementing a broad, economy-wide approach in
this highly complex and rapidly changing space.
Australia’s place in global AI regulation
In the global context of AI regulation, Australia’s position having regard to its size and
GDP, is one of a participant rather than a trendsetter. Given that Australian entities are
or anticipate exporting their AI products or services, alignment or harmonisation with
international regulatory frameworks and limiting additional compliance burdens as far as
reasonably possible is important to ensure Australian organisations can operate
effectively and competitively in a global AI market.
Definition of AI
Against all of these complexities, there is also the persistent issue that the ontological
delineation of AI remains a subject of both profound scholarly contention and practical
contention as demonstrated in the submission writing workshops facilitated by Stirling &
Rose in Sydney and Perth in September 2024.3