- Areas of focus
- In addition to the standard administration and expenditure information provided by agencies, the Committee requested that agencies provide specific additional commentary on the following areas of focus:
- Artificial Intelligence (AI), Machine Learning (ML) and Bio-Intelligence, including governance—being the ethical and responsible management of AI, ML and Bio-Intelligence systems, ensuring that they are developed, deployed, and used in a manner that aligns with societal values and priorities in addition to the Australian Government’s intelligence priorities; how AI, ML and Bio-Intelligence internal systems are assessed against common standards, details of that assessment, common trends and future challenges in this space.
- Detail of policies, guidelines and other initiatives for staff support services; including the provision of psychological support services, as well as support of staff career development in terms of career planning.
- Strategic workforce planning activities and initiatives in development or implemented, aimed at attracting and retaining culturally and linguistically diverse (CALD) staff with native language proficiency.
- The use of external contractors including PwC, Deloitte, KPMG and EY; including the value of contracts (starting value and end value), whether contracts were awarded to an entity that also assisted in any aspect of the tender process, and any initiatives aimed at reducing the use of such contractors.
Artificial Intelligence, Machine Learning and Bio-Intelligence
3.2No agency provided information on use of bio-intelligence. Some agencies stated they did not use any bio-intelligence capabilities.
3.3The Australian Security Intelligence Organisation (ASIO) described itself as a ‘human-led, data-driven, technology enabled organisation ’that has an ongoing and strong focus’ on AI and ML to ensure it maximises ‘the opportunities presented by technological innovations’
3.4During the reporting period ASIO published its Ethical artificial intelligence principles, which outline the seven cornerstone values ASIO applies in developing, maintaining and using AI-based capabilities as follows:
Human-led - Our use of AI supports human decision making. While we find ourselves required to delegate some routine, large-scale tasks to AI due to increasing volumes of data, we ensure that ultimately humans are accountable for decision making and that the use of AI includes an appropriate level of human involvement for the particular use-case. This is especially relevant in the case of decisions significantly affecting an individual.
Data-driven - We actively endeavour to evaluate the accuracy and limitations of AI models and determine whether an AI solution is appropriate for a given problem. We strive to understand, and where possible minimise through technical design and data usage, the biases inherently present in all AI.
Considered - We seek to recognise harms that may arise from the use of AI and act to minimise those harms whether they be individual or societal, reputational or operational. We balance potential risks and costs of using AI against both the benefits of AI and the risks associated with not using AI.
Accountable - We clearly and actively communicate information on proper usage and limitations of AI capabilities to users. Responsibility for proper use of AI is clearly articulated and a system of checks and balances is in place guiding use where appropriate to ensure our actions are reasonable, fair and proportionate.
Responsive - The use of AI is monitored as use-cases emerge or evolve over time—officers should be empowered to highlight any perceived issues with the use of AI. Where appropriate, systems are updated to ensure the use of AI continues to meet its intended purpose.
Security-conscious - Data and AI models are security classified and handled appropriately. This includes considering the potential for compromise of our capabilities arising from improper classification or sharing of data and AI models which is balanced against the need for the transparency afforded by a lower classification where appropriate.
Compliant and responsible - ASIO's use of AI complies with relevant legislation and codes of practice. Further, our use of AI and the data that support it considers not just whether we can do something (because the action is lawful) but whether we should do it—our use of AI must be respectful of the values of our democratic society as well as upholding ASIO values. We strive for transparency in the development, deployment and reporting of AI capabilities, to ensure we maintain our licence to operate.
3.5In January 2023 the Australian Signals Directorate (ASD) released the document Ethical artificial intelligence in the Australian Signals Directorate.In it ASD recognised the following AI ethical principles:
Lawful and appropriate: ASD’s development, management, and deployment of AI capabilities are consistent with the legislation, policies, processes and frameworks that govern ASD’s functions and protect the privacy of Australian citizens.
Enabling human decision-making: Human assessment and judgement will remain central. ASD will use AI to support and enable its highly-skilled workforce to best fulfil ASD’s functions in protecting Australia and Australians from threats.
Reliable and secure: AI capabilities developed by ASD will be managed to ensure that they are reliable, continue to meet their intended purpose, and remain protected from external interference.
Accurate and fair: ASD will endeavour to remove unintended bias from its AI systems so that they produce results that are balanced, accurate and fair.
Accountable, transparent and explainable: AI-based capabilities will have human oversight and control, with clear accountabilities in place for all stages of the development life-cycle. ASD is committed to meeting the ethical need for its decisions and actions to be accountable, transparent and explainable to ensure it operates appropriately and proportionately. This will be balanced with the need to protect sensitive equities to ensure ASD is able to continue to perform its critical intelligence and security functions.
3.6In addition ASD set out some specific challenges it seeks to address through the application of the above principles:
Minimising unintended bias: AI models can reflect biases made by system creators in their training data and design decisions. This can lead to biased results such as disadvantaging certain demographics. Appropriate guidelines and governance arrangements will be developed to ensure that ASD minimises unintended bias and that the true and intended outcome of an AI system aligns with ASD’s licence to operate.
Appropriate use of AI models: Using a model for a purpose for which it was not originally designed can have unintended consequences that impact on individuals. Even when a model is used for its intended purpose, model drift over time may introduce biases or inaccuracy, meaning that it is no longer suitable for its intended purpose. Governance related to the ongoing applicability of AI models and standards for accuracy and reliability will be developed to ensure the continued appropriate use of AI systems in ASD.
Enabling human decision-making: Effective human-machine teaming is important for users to understand how to make informed decisions based on AI system outputs, and to maintain trust in AI systems. ASD’s workforce will have a solid foundational knowledge of how AI works, enabling them to use the right model on the right data and the confidence to judge the level of certainty of the results.
Securing AI systems: Attacks by adversaries could affect the behaviour of AI systems and the decisions made as a result of their outputs. Standards for system monitoring, reliability and life-cycle management will be developed to safeguard the health and security of AI systems.
3.7The Defence Intelligence Organisation (DIO) said that it
continues to explore the potential of a range of commercial off-the-shelf open source tools that have AI/ML enabled capabilities. These tools would enhance the ability of DIO analysts to research, curate and fuse open source information into DIO assessments.
3.8The Australian Geospatial-Intelligence Organisation (AGO), the Office of National Intelligence (ONI) and the Australian Secret Intelligence Service (ASIS) provided classified information on their use of AI and ML.
Staff support services
3.9The agencies were invited to provide detail of policies, guidelines and other initiatives for staff support services; including the provision of psychological support services, as well as support of staff career development in terms of career planning.
Psychological support services
3.10Each of the six intelligence agencies provided a response at various national security classifications in relation to the psychological support services they provide for staff.
3.11ONI stated:
ONI Psychological Services is committed to providing a range of services based on evidence informed best practice to support staff and managers. ONI Psychological Services include mental health briefings, promotion of psychological resilience, self-care training, vocational counselling and career guidance as well as specialist psychological intervention.
3.12Noting that they are both situated within the Defence Intelligence Group (DIG), AGO and DIO stated that they have
an embedded TS Positive Vetting in-house psychological services team to support employees and management , DIG Psychological Services operate with an ‘open door’ policy and provide staff with internal counselling support, as well as counselling referrals to highly experienced psychologists in the community. DIG Psychological Services also offer staff a wide range of services, such as specialised training to staff exposed to sensitive content in their roles, training around wellbeing topics such as resilience and burnout, training and advice to managers, and tailored psychoeducation and debriefing to individuals and groups.
3.13ASD stated that it ‘has a strong focus on mental health and wellbeing which is essential in maintaining a workforce that is emotionally resilient and high-performing’ and that in 2022-23
the Psychological Services Team focused on increasing psychological support to all staff in remote locations and overseas. The team has continued to provide psychological suitability assessments to staff on posting to ensure their wellbeing while representing ASD. The team also continued to provide training for exposure to traumatic material and the Trauma informed Leadership module in order to assist leaders in meeting their WHS obligations. Other offerings included managing uncertainty and resilience training.
3.14ASIO and ASIS provided classified information on this matter.
Support of staff career development
3.15As they are both situated as part of the DIG, DIO and AGO provided near identical information in relation to this area of focus with DIO stating:
Career development for [Australian Public Service] APS and [Australian Defence Force] ADF staff is supported by the DIG Workforce Plan 2023-26 and Implementation Plan (the Plans) which together identify three objectives aimed at improving career development and planning initiatives. The Plans will clarify current APS skills and capability requirements to ensure alignment with current and future workforce demands, and will subsequently establish suitable career pathways. The DIG is also improving career development services for ADF staff through engagement with senior manager, implementation of the Defence People System, and support to leaders and managers to better articulate the role requirements, expectations and opportunities available to the ADF members that report to them.
3.16Other agencies provided information on leadership skills development and more general skills development as well as other training but did not specifically address career development.
Attracting and retaining culturally and linguistically diverse (CALD) staff with native language proficiency
3.17ASIO and ASIS provided classified evidence on this issue.
3.18ONI stated that through its CALD network it
played a significant leadership role in recognising and better engaging with culturally and linguistically diverse staff from across the NIC. The ONI CALD Network brought together NIC CALD Networks form ASD, ASIO, ASIS, Defence Intelligence Organisation (DIO), Australian Geospatial Organisation (AGO) at the SES Champion level, but also amongst NIC staff, culminating in the first in-person interagency Harmony Day event held at the Ben Chifley Building.
The event featured a CALD panel discussion involving four NIC agencies (ASD, ASIO, ASIS and ONI) exploring the experience, perspectives and issues of CALD staff working in the NIC and opportunities to improve inclusive practice in recruitment, recognition, retention and representation; improving pathways to progression to leadership; and changing the narrative and embedding diversity and inclusion in NIC organisational culture. ONI also worked with the National Security College on the delivery of a CALD podcast mini-series to develop a public narrative around CALD, focussed on raising and exploring CALD issues in the national security community, and used the opportunity to raise awareness of opportunities for people of CALD background to work in the NIC.
3.19DIO and AGO provided consistent information in relation to this area of focus referring to graduate recruitment activities within the DIG. Further, AGO informed the Committee that, during the reporting period, DIG
established a Group-level Culture and Diversity Advisory Group to advise on cultural and behavioural matters to inform and affect Group-level change. Through the Advisory Group, we track and participate in key culture events and prioritise diversity and inclusion initiatives to make DIG an inclusive and welcoming place to work. Initiatives to date have focused on attraction and retention of staff, considering neurodiversity, cultural and linguistic diversity, gender, sexuality and more.
3.20DIO stated that the modernisation of its workforce
allows the attraction and retention of CALD staff with native language proficiency through actions targeting revised recruitment strategies, job family alignment, specialist training and cultural reform initiatives.
Use of external contractors
3.21ASIO explained that it engages consultants where it does not have ‘in-house capability’ and that it has
commenced an organisational wide contract rationalisation/optimisation exercise. It will seek to identify, assess and evaluate existing contracts across the enterprise and catalogue them according to their requirement category (i.e. ICT, facility etc.) and value. Duplication will be identified and reduced through implementation of enterprise-wide agreements.
3.22ONI reported that it had contracts with PWC and KPMG during the period. It stated that contractors
are engaged for short term projects or to fill APS positions while undertaking recruitment activity. No contracts were awarded to an entity that also assisted in any aspect of the tender process.
3.23ASD had a number of contracts with KPMG, PWC, Deloitte and EY. It explained that it
primarily engages contractors through the use of available panel arrangements established through an open tender process. In some limited instances direct contracting is adopted where an essential precondition to effectively perform a particular contract has been identified. No contract has been awarded to an entity that assisted in any aspect of the tender process.
3.24DIO stated that it did not use contractors from PWC, Deloitte, KPMG or EY and that it is ‘committed to actively reducing contractor hours or terminating contracts, where possible’.
3.25AGO reported that it had five commercial arrangements with KPMG during the reporting period. The value of the contracts totalled $4.967 million with $1.434 million expensed during the reporting period. Of the five contracts three were active at 30 June 2023.
3.26ASIS provided classified information on this issue.
Committee Comment
3.27Overall, the Committee was satisfied with the intelligence agencies’ evidence on the areas of focus for this inquiry though some improvement could be made. Due to the differing levels of national security classification at which information is provided some agencies provided detailed information that is not able to be included in this report, and the quality of evidence varied between the agencies and between the focus areas.
3.28In particular the Committee notes its request for focussed information on: Detail of policies, guidelines and other initiatives for staff support services; including the provision of psychological support services, as well as support of staff career development in terms of career planning and notes that this was mostly addressed in the terms quoted above, while no agency provided actual policies or guidelines in relation to this term of reference. In the Committee’s view simply stating that an agency has a psychologist who is available to discuss certain issues is not sufficient to assure the Committee that the psychological needs of staff are being adequately identified and addressed.
3.29Some agencies did not provide any separate detail in relation to staff career development in terms of career planning. Whilst they did provide information on leadership skills development and more general skills development as well as other training, which can be assumed to assist staff in their career development, it should not be left to the Committee to assume career development is a priority for an agency merely because some aspects of the agency’s training and development can broadly be linked to that outcome.
3.30The Committee was pleased with agencies’ engagement with the issue of attracting and retaining CALD staff with native language proficiency. In particular the Committee notes the significant work of ONI through its CALD network and other agencies’ focus on cultural diversity in recruitment and retention practices.
3.31In relation to AI, the Committee notes the following from the 2020 Comprehensive Review of theLegal Framework of the National Intelligence Community by Mr Dennis Richardson AC:
3.96 It would be premature to legislate to control the use of artificial intelligence for intelligence purposes. Many of the NIC agencies are beginning to develop and implement artificial intelligence capabilities to support their intelligence function. However, as at December 2019, these capabilities are in their infancy—they are not yet at a point where the Government could usefully or sensibly introduce controls or safeguards. No comparable Western country has introduced statutory controls on the use of artificial intelligence for intelligence purposes.
3.97 However, it is clearly foreseeable that a time will come, when it will be necessary to introduce statutory controls. As agencies’ capabilities become more advanced, they will increasingly have the potential to raise complex legal and ethical issues that may require a legislative response—most significantly, the potential for advanced artificial intelligence capabilities to be used to make far-reaching decisions about individuals and/or to engage in mass surveillance. In the meantime, the requirement to have human involvement in significant or adverse decisions made by automated capabilities or artificial intelligence should be maintained.
3.98 We recommend that the Parliamentary Joint Committee on Intelligence and Security (PJCIS) should be briefed annually on agencies’ development of artificial intelligence capabilities, and that future Independent Intelligence Reviews should consider whether further policy or legislative responses are required.
3.32The Government Response to the above stated:
Recommendation 159: The PJCIS should receive a briefing from agencies on the development of their artificial intelligence-based intelligence capabilities at least once per year.
Agreed. The purpose of these briefings should be to enable the Parliament to develop and maintain an understanding of NIC agencies’ development and use of AI. Agencies should inform the Parliamentary Joint Committee on Intelligence and Security (PJCIS), within the scope of its existing functions, of any substantial changes in their development and use of AI capabilities, to provide Parliament with an opportunity to consider whether legislative or policy responses are required. In such briefings, the PJCIS should also consider whether the IGIS has access to sufficient independent technical expertise necessary to scrutinise NIC agencies’ use of AI capabilities for intelligence purposes.
3.33The Committee notes the Government’s agreement that it should be briefed by agencies on the development of their AI-based intelligence capabilities at least once per year. It further notes that this review’s focus on AI/ML has been well responded to by agencies and the Committee looks forward to further engagement with agencies on AI/ML.
3.34The Committee acknowledges that out of all the focus areas, the specific information provided in relation to AI/ML was the most detailed and useful, clearly setting out agencies’ engagement with, thinking about and planning for the implementation and use of these complex technological tools. The Committee welcomed ASIO’s additional provision of a specific in-house briefing to deepen the Committee’s understanding of its use of AI, and relevant considerations about technology, governance and policy.
3.35Some of the information provided to the Committee on the use of external contractors was detailed and, when asked, further detail was readily provided by agencies. The Committee notes some agencies’ efforts to review and rationalise the use of external consultants, in line with the Whole-of-Government approach to the reduction in the use of consultants in the APS.
3.36Overall, the Committee is pleased with the level of use of consultancies by the NIC and encourages agencies to continue their endeavours and their transparency in this regard. The Committee believes that reduction in their use, will prompt the NIC to become more self-reliant, efficient, and adaptable in today’s evolving and complex security environment.
Senator Raff Ciccone
Chair
27 November 2024