Australia - Victoria - AI Recordkeeping Policy (RAI-AU-VI-AIATRXX-2024)

Overview

The Artificial Intelligence (AI) Technologies and Recordkeeping Policy, published by the Public Record Office Victoria (PROV), serves as a foundational document for all Victorian public offices concerning the judicious and compliant integration of artificial intelligence technologies within their operational frameworks. This policy was meticulously crafted to address the burgeoning complexities and unique challenges that AI systems introduce to traditional recordkeeping practices. Its primary objective is to ensure that the deployment and utilization of AI technologies within the public sector uphold the fundamental principles of transparency, accountability, and explainability, which are paramount for maintaining public trust and operational integrity. By providing clear directives, the policy enables public offices to navigate the intricate landscape of AI implementation while adhering strictly to their statutory obligations under the Public Records Act 1973. The policy underscores that records generated by or through AI technologies are indeed public records, necessitating their management in full accordance with PROV Recordkeeping Standards.

Furthermore, the policy is designed to be adaptable, acknowledging the diverse applications and varying risk profiles associated with different AI technologies. It provides a framework that allows public offices to tailor their documentation and recordkeeping approaches based on the specific context of AI implementation and the potential for harm. This nuanced approach ensures that the policy is not overly prescriptive but rather empowers agencies to implement robust recordkeeping practices proportionate to the risks involved. It emphasizes the importance of documenting not only the outputs of AI systems but also the underlying processes, data sources, and human interventions, thereby creating a comprehensive audit trail. This holistic view of recordkeeping is crucial for fostering an environment where AI technologies can be leveraged for efficiency and innovation without compromising ethical considerations or legal compliance.

Definitions

While the policy itself does not contain a dedicated, exhaustive glossary section, it implicitly defines or elaborates upon several key concepts central to its application. 'Artificial Intelligence (AI) Technologies' are understood broadly to encompass systems that can undertake tasks in line with directives, often involving automation, machine learning, and related technologies that may make or inform decisions. The policy's scope extends to any technology that generates or influences records through automated or semi-automated processes. 'Public Records' are reaffirmed as any information created, received, or maintained by a public office in the course of its business, explicitly including content created by or through the use of AI technologies. This clarification is fundamental, as it establishes the legal obligation for AI-generated content to be managed under existing recordkeeping standards.

The concept of 'Recordkeeping Standards' refers to the established guidelines and requirements issued by the Public Record Office Victoria, which dictate how public records must be created, captured, managed, and disposed of. The policy mandates that AI-generated records must conform to these standards, ensuring their authenticity, reliability, integrity, and usability over time. 'Transparency' and 'Accountability' are central tenets, implying that the processes, inputs, and outputs of AI systems must be sufficiently documented and understandable to allow for scrutiny and attribution of responsibility. 'Explainable AI' is a critical objective, meaning that the rationale and mechanisms behind AI-driven decisions or content generation should be discernible, particularly when these have significant impacts on individuals or public services. These implicit definitions form the bedrock upon which public offices are expected to build their AI recordkeeping strategies, ensuring consistency and compliance across the Victorian public sector.

Governance and Institutional Framework

The governance framework for AI recordkeeping in Victoria is primarily anchored by the Public Record Office Victoria (PROV), which is responsible for issuing the Artificial Intelligence (AI) Technologies and Recordkeeping Policy and providing ongoing guidance to public offices. PROV's role extends to setting the overarching recordkeeping standards that all Victorian public sector entities must adhere to, ensuring consistency and compliance with the Public Records Act 1973. The policy explicitly places the ultimate responsibility for the creation and management of full and accurate records, including those generated by or through AI technologies, with the head of the respective public office. This clear delineation of responsibility ensures that accountability for AI-driven recordkeeping practices is firmly established at the highest levels of each agency. Public offices are expected to integrate the policy's directives into their existing governance structures, ensuring that AI implementation is overseen by appropriate internal committees or designated personnel.

The institutional framework also mandates a collaborative approach, where public offices must consider the policy in conjunction with other relevant PROV policies, such as the Approval Processes Policy and the PROV Value and Risk Policy, as well as broader recordkeeping standards. This integrated approach ensures that AI recordkeeping is not treated in isolation but as an integral component of an agency's overall information governance strategy. Furthermore, the policy encourages public offices to engage with other relevant bodies for advice and guidance, including the Office of the Victorian Information Commissioner (OVIC) for privacy and data protection matters, and the Victorian Equal Opportunity and Human Rights Commission for issues related to the Charter of Human Rights. This multi-faceted governance structure aims to provide comprehensive oversight and support for the responsible and ethical deployment of AI technologies across the Victorian public sector, ensuring that recordkeeping obligations are met while fostering innovation.

Key Focus Areas

The Victoria AI Recordkeeping Policy delineates several critical focus areas to ensure comprehensive and compliant management of AI-generated records. A primary focus is on the documentation of AI technologies, requiring public offices to maintain detailed records about why and how specific AI technologies are being used, the types of records they generate, the location of source data, and any data or processes that should be excluded from AI access. This extensive documentation is crucial for establishing an audit trail and enabling future scrutiny or explanation of AI system operations. The policy emphasizes that the level of documentation should be proportionate to the context of AI implementation and the level of risk involved, allowing for flexibility while ensuring sufficient detail for accountability.

Another significant focus area is human oversight and intervention. The policy explicitly states that processes to check and confirm the accuracy of records and data created by or through AI must be overseen by a human being. This directive underscores the importance of human accountability in validating AI outputs, mitigating potential errors, biases, or unintended consequences. Similarly, the disposal of public records, even those generated by AI, must be overseen and approved by a human being, reinforcing the principle that critical recordkeeping decisions remain within human control. This human-in-the-loop approach is fundamental to ensuring the reliability and integrity of public records in an AI-driven environment. The policy also highlights the need for transparency and responsible disclosure, so individuals can understand when they are significantly impacted by AI and can ascertain when an AI system is engaging with them, thereby promoting explainable AI use.

Implementation Framework

The implementation framework for the Artificial Intelligence (AI) Technologies and Recordkeeping Policy requires Victorian public offices to integrate its directives into their existing recordkeeping and information governance practices. Public offices are mandated to identify, describe, and document each AI system deployed across their organisation. This documentation should comprehensively cover the intended purpose and desired outcomes of the AI system, the specific type of AI technology being utilised, its operational context and scope of use, and its current stage of implementation. Furthermore, agencies must meticulously record the sources of data feeding into the AI system, any identified harms and risks associated with its use, and the controls or mitigation strategies put in place to address these risks. This detailed mapping of AI systems is crucial for establishing a clear understanding of their impact on record creation and management.

A key aspect of the implementation framework involves determining and documenting what content can be created and captured by AI technologies versus what necessitates human creation and capture. For AI-generated content, public offices must clearly define at what points human intervention is required for review, confirmation, or direct action. This extends to actions undertaken by AI technologies, where the framework dictates that human beings must be included to confirm or undertake actions, ensuring appropriate oversight. The policy also stresses the importance of transparency and accountability regarding ownership and responsibility for all types of AI used, ensuring that individuals responsible for different phases of the AI system lifecycle are identifiable and accountable for the system's outcomes. This comprehensive approach to implementation ensures that AI technologies are deployed and managed in a manner that upholds recordkeeping integrity and public accountability.

Monitoring and Evaluation

The Public Record Office Victoria's Artificial Intelligence (AI) Technologies and Recordkeeping Policy implicitly necessitates robust mechanisms for monitoring and evaluating the effectiveness of AI recordkeeping practices within public offices. While the policy itself does not outline a specific, standalone monitoring and evaluation program, its directives inherently require ongoing assessment to ensure compliance and adapt to evolving AI landscapes. Public offices are expected to establish internal processes for regularly reviewing their AI systems' documentation, ensuring that it remains accurate, up-to-date, and reflective of the actual operation and impact of the AI technologies. This includes periodic checks on the types of records being generated, the integrity of source data, and the efficacy of human oversight mechanisms. The dynamic nature of AI development and deployment means that monitoring cannot be a one-off event but rather an continuous process to identify new risks, assess the effectiveness of mitigation strategies, and ensure ongoing adherence to recordkeeping standards.

Furthermore, the policy's emphasis on transparency and accountability naturally extends to the need for evaluating whether AI systems are indeed explainable and if their outputs are reliable and unbiased. Public offices should consider implementing audit trails and logging mechanisms that capture AI system activities, decisions, and any human interventions. This data can then be used to evaluate the system's performance against its intended purpose and to identify any discrepancies or anomalies in record creation. The policy also encourages public offices to align their practices with broader national frameworks, such as the National Framework for the Assurance of Artificial Intelligence in Government, which inherently promotes continuous assurance and evaluation. By fostering a culture of continuous monitoring and evaluation, public offices can ensure that their use of AI technologies remains compliant with recordkeeping obligations, ethical principles, and public expectations for transparency and accountability.

Penalties, Liability, and Appeals

The Artificial Intelligence (AI) Technologies and Recordkeeping Policy itself does not explicitly detail specific penalties, liability provisions, or appeal mechanisms directly related to non-compliance with its directives. Instead, the policy operates within the existing legal framework established by the Public Records Act 1973 (Vic), which governs recordkeeping obligations for all Victorian public offices. Therefore, any failure to adhere to the AI Recordkeeping Policy's requirements would likely fall under the broader enforcement provisions of the Public Records Act. This Act outlines the responsibilities of public offices and their heads regarding the creation, maintenance, and disposal of public records, and non-compliance can lead to various consequences, including administrative actions, reputational damage, and potential legal ramifications for breaches of statutory duties. The policy reinforces that records created by or through AI technologies are public records and must be managed in accordance with PROV Standards, thereby extending the existing legal obligations to the AI context.

In terms of liability, the policy places ultimate responsibility for the creation and management of full and accurate records, including AI-generated content, with the head of the public office. This means that accountability for ensuring compliance rests at a senior level within each agency. While the policy emphasizes human oversight and intervention in AI processes, it does not create new specific liability regimes for AI-related recordkeeping failures. Instead, it integrates AI recordkeeping into the established legal framework where existing principles of administrative law, duty of care, and public sector accountability would apply. For appeals, any grievances or disputes related to recordkeeping decisions, including those involving AI, would typically be handled through existing administrative review processes or, in more serious cases, judicial review, as provided for under Victorian law. The policy's strength lies in its integration with, and reinforcement of, existing legal obligations rather than establishing a separate enforcement mechanism.

Relationship to Other Instruments

The Artificial Intelligence (AI) Technologies and Recordkeeping Policy is not a standalone document but is intricately woven into a broader ecosystem of legal and policy instruments within the Victorian government. It explicitly functions as a companion to several other key policies issued by the Public Record Office Victoria (PROV), including the Approval Processes Policy, which covers decisions and actions relating to approval processes, including those conducted by machine learning and related technologies. This ensures a consistent approach to recordkeeping across automated decision-making. Furthermore, public offices are directed to apply the terms of the AI Recordkeeping Policy in line with the PROV Value and Risk Policy and the overarching Recordkeeping Standards, which collectively provide a comprehensive framework for managing public records. This interconnectedness ensures that AI recordkeeping is integrated into a holistic information governance strategy, rather than being treated as an isolated concern.

Beyond PROV's internal policy suite, the AI Recordkeeping Policy operates within the context of several foundational Victorian and Australian laws. It is designed to aid public offices in meeting their statutory obligations under the Public Records Act 1973 (Vic), which is the primary legislation governing public recordkeeping in the state. The policy also has strong linkages to privacy legislation, particularly the Privacy and Data Protection Act 2014 (Vic), as the use of AI often involves the processing of personal information. Public offices are also advised to consider their obligations under the Victorian Charter of Human Rights and Responsibilities Act 2006 and the Public Administration Act 2004. Additionally, the policy acknowledges and aligns with the National Framework for the Assurance of Artificial Intelligence in Government, which Victoria agreed to in June 2024, demonstrating an effort towards national consistency and the application of Australia's AI Ethics Principles. This layered approach ensures that AI recordkeeping adheres to a wide range of legal, ethical, and administrative requirements.

International Alignment

The Victoria AI Recordkeeping Policy demonstrates an awareness of broader national and, by extension, international discussions surrounding artificial intelligence governance, particularly through its alignment with the National Framework for the Assurance of Artificial Intelligence in Government. Victoria's agreement to this National Framework in June 2024 signifies a commitment to a nationally consistent approach to assuring AI in government, which in turn assists in applying Australia's AI Ethics Principles. While the policy itself does not directly reference international treaties or specific foreign regulations, its adherence to principles such as transparency, accountability, and explainability for AI systems reflects a global consensus on responsible AI development and deployment. These principles are echoed in various international guidelines and recommendations from bodies like the OECD and UNESCO, which advocate for human-centric and trustworthy AI.

By integrating with a national framework that promotes ethical AI principles, the Victorian policy indirectly contributes to international best practices. The emphasis on human oversight, risk assessment, and the creation of auditable records for AI systems aligns with the broader international push for robust governance mechanisms to address the societal impacts of AI. Although the policy's primary focus is on domestic recordkeeping obligations, its foundational principles are consistent with the evolving global dialogue on AI regulation, which prioritizes ethical considerations, data protection, and the need for clear accountability when AI systems are used in public service. This alignment ensures that Victoria's approach to AI recordkeeping is not insular but rather contributes to and benefits from wider efforts to establish responsible AI governance.

Implementation Timeline

Compliance Checklist

Sources and References