Item Catalogue
AI governance, regulation, strategy, and practice developments from monitored sources.
- Week of 11 May 2026
-
Musk v. Altman week 3: Elon Musk and Sam Altman traded blows over each other’s credibility. Now the jury will pick a side.
- The Musk v. Altman trial entered closing arguments, with credibility and nonprofit commitments as central issues.
- Altman faced scrutiny over conflicts of interest; a US House committee and state AGs launched separate investigations.
- This is a US civil litigation item with negligible direct relevance to Australian federal AI governance.
- Week of 4 May 2026
-
National AI Centre launches AI.gov.au
- DISR's National AI Centre has launched AI.gov.au, centralising practical AI guidance, tools, and resources for Australian organisations.
- The platform targets SMEs and not-for-profits but consolidates content relevant to APS agencies developing their own guidance.
- AI.gov.au will support the AI Safety Institute by making safety guidance more accessible to smaller organisations over time.
-
Judge Finds DOGE Used ChatGPT to Cancel Grants
- A US federal judge ruled DOGE's ChatGPT-assisted grant cancellations unconstitutional, overturning 1,400+ terminations worth $100M.
- The case documents how minimal-context LLM prompts without human-in-the-loop review produced legally invalid government decisions.
- Directly relevant to APS agencies considering AI-assisted screening, eligibility, or allocation decisions under Australian administrative law.
-
APRA Warns Risk Management Trails Rapid A.I. Adoption
- APRA warns that governance and risk management are not keeping pace with AI adoption in financial services.
- APRA's late-2025 targeted engagement found boards lack AI technical literacy and over-rely on vendor presentations.
- APRA calls for a 'step-change' in AI risk management and sets minimum board oversight expectations.
-
New Zealand frames non-binding AI guidance for government
- New Zealand has published a voluntary, non-binding AI framework for its public sector naming transparency, fairness, and human oversight.
- Academic authors criticise the approach as 'Pollyanna policy', noting binding frameworks produce stronger audit and procurement outcomes.
- Australia faces a closely analogous policy design question - voluntary versus binding AI governance for government agencies.
-
White House Considers Pre-Release Vetting for AI Models
- The White House is actively deliberating an executive order requiring government pre-release vetting of frontier AI models.
- Anthropic's Mythos model - capable of finding and exploiting software vulnerabilities - is cited as a proximate trigger for the policy shift.
- No formal order has been issued; the White House described current reporting as speculation, limiting immediate policy impact.
-
CAISI Signs Agreements Regarding Frontier AI National Security Testing With Google DeepMind, Microsoft and xAI
- CAISI formalises pre-deployment and post-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI.
- Evaluations include models with reduced safeguards, conducted in classified environments via an interagency taskforce.
- This US model of government-led frontier AI safety testing may inform expectations placed on Australia's AISI.
-
Import AI 455: AI systems are about to start building themselves.
- Jack Clark estimates 60%+ probability of fully automated AI R&D—no human involvement—by end of 2028.
- Benchmark evidence shows AI task horizons expanding from 30 seconds in 2022 to 12 hours in 2026, with 100-hour tasks projected by end of 2026.
- The analysis is expert opinion and probabilistic forecasting, not confirmed policy or technical fact—treat as informed signal, not settled consensus.
-
Third GPAI Signatory Taskforce meeting – Safety and Security chapter
- The EU GPAI Code of Practice Taskforce is developing structured aggregate risk forecasting for frontier AI providers with systemic risk.
- Proposed semi-annual forecasting exercises would aggregate anonymised individual provider estimates into industry-wide risk signals.
- Directly relevant to Australian AISI and DISR work on frontier AI safety given Australia's GPAI membership and international alignment.
-
Commission opens consultation on draft guidelines for AI transparency obligations
- EU AI Act transparency obligations take effect 2 August 2026, requiring disclosure when people interact with AI or AI-generated content.
- Draft guidelines clarify scope for providers and deployers; a voluntary Code of Practice on AI content marking is expected June 2026.
- Australian agencies with EU-facing services or procuring EU-developed AI systems may need to assess compliance exposure.
-
AI Governance in Practice: Trusted AI in Age Verification Systems
- KJR podcast explores AI governance lessons drawn from Australia's Age Assurance Technology Trial, where KJR was Test & Evaluation Partner.
- Age verification use case illustrates broader AI governance challenges: risk-tiered accuracy, hallucinations, and system-specific governance frameworks.
- Content is practitioner-oriented and vendor-authored; useful context but not independent research or authoritative guidance.
-
EU agrees to simplify AI rules to boost innovation and ban ‘nudification' apps to protect citizens
- EU Parliament and Council agree to simplify AI Act implementation, easing compliance burdens for businesses.
- High-risk AI rules (biometrics, employment, migration) now apply from December 2027, not earlier dates.
- Australia's own AI governance is not directly bound, but the EU's sequencing approach offers a comparable model.
-
CEOs Expect AI to Make 48% of Operational Decisions by 2030
- IBM survey of 2,000 CEOs finds 48% of operational decisions expected to be AI-made without human input by 2030.
- Chief AI Officer appointments jumped from 26% to 76% of organisations in one year, signalling rapid executive accountability shifts.
- Only 25% of employees currently use AI regularly, pointing to a significant workforce readiness gap ahead of projected automation.
-
Google launches Gemini Enterprise Agent Platform for governance
- Google's Gemini Enterprise Agent Platform consolidates agent building, deployment, and governance into one environment.
- Built-in governance primitives include an Agent Registry, audit logs, semantic policies, and identity controls via Agent Gateway.
- Vendor-supplied governance lowers engineering overhead but does not substitute for agency-level policy mapping and validation work.
-
Cybersecurity and AI: The Evolving Security Landscape
- AI is expected to automate the full cyberattack chain, lowering barriers for non-state actors targeting critical infrastructure.
- Defensive AI gains are undermined by persistent failures in patch management, configuration, and security hygiene across operators.
- Analysis is US-focused and undated; Australian critical infrastructure context adds relevance but is not directly addressed.
-
The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
- CAIS releases a 4,157-question benchmark measuring hazardous WMD-related knowledge in LLMs across bio, cyber, and chemical domains.
- A new 'unlearning' method (CUT) removes hazardous knowledge from models entirely, making jailbreaking ineffective at eliciting it.
- Dual-use knowledge can be preserved for approved professionals via structured API access, offering a nuanced safety model.
-
U.S., China Weigh Bilateral AI Guardrails Talks
- Washington and Beijing are reportedly considering placing AI governance on the agenda of an upcoming Trump-Xi summit.
- Bilateral AI talks between major powers typically precede shifts in export controls, chip access, and cross-border compliance requirements.
- Reporting is early-stage and based on unnamed sources; no formal agenda or outcomes have been confirmed.
-
Microsoft removes Copilot branding from Windows 11 apps
- Microsoft is removing Copilot branding from Windows 11 apps while retaining the underlying AI functionality.
- A new RemoveMicrosoftCopilotApp Group Policy lets enterprise admins uninstall Copilot from managed Windows 11 devices post-April 2026.
- Changes affect Enterprise, Professional and Education SKUs - the editions most common in APS desktop environments.
-
NIST NCCoE Cyber AI Profile Virtual Working Session Series: Extending the Technical Content
- NIST NCCoE is running a virtual working series to develop the Cyber AI Profile under the Cybersecurity Framework.
- Session 2 focuses on extending technical content to cover Agentic AI and Zero Trust architectures.
- Open to government participants - APS cyber and AI governance teams could contribute or monitor outputs.
-
AI Safety, Ethics, and Society
- Centre for AI Safety has released a free interdisciplinary textbook and online course on AI safety, ethics, and governance.
- Covers technical safety, ethics, game theory, complex systems, and AI governance - relevant to APS capability uplift.
- The online course ran July–October 2024; the item is undated and may now be past its enrolment window.
-
Biosecurity and AI: Risks and Opportunities
- AI could lower barriers to bioterrorism by coaching users through viral synthesis and attack planning.
- The analysis proposes targeted controls on AI biotech tools, DNA synthesis access, and sequence screening.
- APS-relevant as biosecurity intersects with AI governance, but this is a think-tank blog without direct Australian policy application.
-
Inside the AI Index: 12 Takeaways from the 2026 Report
- Stanford HAI's 2026 AI Index documents breakthrough AI capabilities alongside growing concerns about environmental costs and transparency.
- The annual AI Index is a widely cited benchmark used by governments and agencies to frame AI strategy and policy positions.
- Extracted text is a teaser only - substantive findings are not available from this item as published.
-
A blueprint for using AI to strengthen democracy
- Personal AI agents that research, lobby, and act on users' behalf risk fundamentally mediating citizen-institution relationships.
- Collective agent interactions could produce democratic harms even when individual agents are well-aligned with their users.
- Primarily a conceptual essay for a general audience; limited direct policy prescription for APS practitioners.
-
IAPP Executive Describes Who Owns AI Governance
- IAPP research finds no consistent model for AI governance ownership across organisations, with privacy teams often absorbing responsibility.
- Cybersecurity, data-governance, and privacy functions are all being pulled into AI governance work, creating coordination friction.
- This is a private-sector-focused interview with limited direct APS applicability, though the governance fragmentation pattern is familiar.
-
Representation Engineering: a New Way of Understanding Models
- Representation engineering identifies and manipulates internal AI activations to detect traits like honesty and power-seeking.
- The technique enables real-time detection and control of model behaviour - relevant to AI assurance and transparency efforts.
- This is foundational research from CAIS; practical application in government AI assurance contexts remains early-stage.