Weekly Digest
Week of 23 Feb 2026
This week at a glance
This week's digest centres on procurement, measurement, and harm classification as practical anchors for AI governance work. The Digital Transformation Agency's new five-year Microsoft agreement (VSA6), commencing July 2026, is the most directly actionable item for APS practitioners, consolidating AI and cloud procurement terms, capping price increases, and strengthening data protections—smaller agencies in particular should note the standardised contracting benefits. On the analytical side, Jacob Steinhardt's argument that technical measurement infrastructure is the most tractable near-term governance intervention sits alongside the MIT-spotlighted EPIC harm taxonomy, together offering practitioners a paired lens for both evaluating AI systems and categorising their potential impacts. The OECD's extended 20 March deadline for AI-in-government contributions is worth noting for agencies with mature implementation or governance examples they may wish to put forward.
Headlines
- AU Gov · DTA signs new 5-year agreement with Microsoft: Delivering value and innovation for the Australian Government
- Global · Deadline extension 20 March: Global call for ‘Governing with Artificial Intelligence’: Share your initiatives and insights on AI-driven innovation in government
- Risk · Generating Harms: Generative AI’s Impact and Paths Forward
- Tech · Import AI 446: Nuclear LLMs; China's big AI benchmark; measurement and AI policy
Australian Government1 item
DTA signs new 5-year agreement with Microsoft: Delivering value and innovation for the Australian Government
The Digital Transformation Agency has secured a new five-year Volume Sourcing Agreement (VSA6) with Microsoft on behalf of the Australian Government, effective 1 July 2026. The agreement provides stable pricing with capped increases, improved discounts, and a standardised contracting framework covering governance, security, liability, and data handling. Notably, it includes provisions to accelerate APS adoption of AI and emerging technologies. The arrangement builds on findings from a recent review of Single Seller Arrangements, which found they delivered $1.6 billion in discounts between 2019 and 2024 and reduced procurement complexity for agencies, particularly smaller entities.
Key points
- DTA has signed a new five-year Volume Sourcing Agreement with Microsoft, commencing 1 July 2026.
- The agreement explicitly accelerates APS capability to adopt AI and other emerging technologies across government.
- Enhanced legal provisions cover governance, security, liability, and handling of government data under the new framework.
Implications
- Monitor Agencies could monitor DTA communications for transition guidance ahead of the 1 July 2026 commencement date.
- Consider AI governance and strategy teams could consider how the new agreement's legal provisions around data handling and AI capability access affect their agency's AI use case planning and risk assessments.
- Consider Procurement and ICT teams in smaller agencies may want to consider engaging with DTA early to understand how the standardised framework simplifies their Microsoft licensing arrangements.
Global Regulation & Policy1 item
Deadline extension 20 March: Global call for ‘Governing with Artificial Intelligence’: Share your initiatives and insights on AI-driven innovation in government
The OECD has issued a global call for submissions under a 'Governing with Artificial Intelligence' initiative, with a deadline extended to 20 March 2026. Governments are invited to share AI use cases, policy initiatives, and implementation tools relevant to trustworthy AI in public administration. The call appears designed to build a comparative evidence base that could inform OECD guidance and peer-learning outputs. The extracted content is limited, so the full scope, selection criteria, and intended publications are not confirmed.
Key points
- OECD invites governments to submit AI use cases, policy initiatives, and implementation tools by 20 March 2026.
- Australian agencies could contribute examples of AI governance practice, potentially shaping OECD comparative outputs.
- Extracted content is brief; full submission scope and intended outputs are unclear from available text.
Implications
- Consider Agencies with mature AI governance or use-case examples may want to consider whether contributing to the OECD call aligns with their international engagement objectives.
- Monitor Policy teams may want to monitor outputs from this OECD initiative as a source of comparative government AI practice and potential benchmarking material.
Risk, Assurance & Ethics1 item
Generating Harms: Generative AI’s Impact and Paths Forward
The MIT AI Risk Repository has spotlighted EPIC's 2023 paper 'Generating Harms', which maps nine categories of harm from generative AI - spanning physical, economic, reputational, psychological, autonomy, discrimination, relationship, opportunity, and dignitary harms - alongside documented real-world cases including deepfakes, defamation, misinformation, and data breaches. The paper also surveys legal, regulatory, and industry interventions. This is a secondary summary from the MIT blog; the underlying EPIC paper is the substantive source. The taxonomy is broadly applicable but originates from a US privacy advocacy perspective and is not calibrated to Australian frameworks such as the APS Responsible AI Policy.
Key points
- EPIC's 2023 framework identifies nine harm categories from generative AI, from physical injury to dignitary harm.
- The MIT AI Risk Repository has catalogued this as its 31st AI risk framework - a growing reference library for governance practitioners.
- Framework is US-origin and advocacy-driven; useful for taxonomy comparison but not directly calibrated to Australian regulatory context.
Implications
- Consider AI governance and risk teams could compare EPIC's nine-category harm taxonomy against their agency's existing AI risk registers or harm assessment frameworks to identify gaps.
- Monitor Teams tracking international AI risk classification approaches may want to note the MIT AI Risk Repository as a growing reference library of curated frameworks.
Technical Developments1 item
Import AI 446: Nuclear LLMs; China's big AI benchmark; measurement and AI policy
Import AI issue 446 covers four distinct research topics. First, Jacob Steinhardt's blog makes the case that investing in AI measurement tools - compute accounting, agent evaluation, privacy-preserving audit - is one of the highest-leverage policy interventions available, noting the field is talent-constrained in this specific area. Second, a King's College London study finds that frontier LLMs (GPT-5.2, Claude Sonnet 4, Gemini 3 Flash) escalate to nuclear use in nearly all simulated wargames and never select de-escalatory options, raising questions about AI advisory dynamics in high-stakes decisions. Third, Chinese institutions have published ForesightSafety Bench, a large-scale AI safety evaluation framework covering categories - including alignment faking, deception, and existential risk - closely mirroring Western frameworks. Fourth, LABBench2 assesses AI capabilities for biological science tasks, finding uneven performance and identifying specific gaps. The issue closes with a speculative fiction piece.
Key points
- Jacob Steinhardt's blog argues measurement infrastructure is a prerequisite for effective AI governance and policy intervention.
- A King's College London study finds LLMs escalate to nuclear use more readily than humans in wargame simulations.
- China's ForesightSafety Bench covers existential-risk and alignment categories similar to Western AI safety evaluation frameworks.
Implications
- Consider Agencies developing AI assurance or governance frameworks may want to consider Steinhardt's measurement-first argument when scoping evaluation capability investments.
- Monitor Policy teams tracking international AI safety benchmarking efforts may want to monitor ForesightSafety Bench as a signal of convergence between Chinese and Western AI safety evaluation norms.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.