Item Catalogue
AI governance, regulation, strategy, and practice developments from monitored sources.
- Week of 3 February 2025
-
LLMs have been set their toughest test yet. What happens when they beat it?
- The Alan Turing Institute blog examines 'Humanity's Last Exam', a new benchmark designed to stress-test frontier LLMs.
- The piece raises what comes after benchmark saturation - a recurring challenge for AI evaluation and assurance frameworks.
- Extracted text is very thin; substantive analysis is unavailable from the provided content alone.
- Week of 27 January 2025
-
NIST Researcher Describes Data Considerations for Industrial Artificial Intelligence
- NIST's second IAI blog instalment covers data quality requirements for industrial AI applications in manufacturing.
- Highlights data pitfalls - incomplete data, inadequate variation, large gaps - relevant to any AI deployment context.
- Beginner-level manufacturing focus limits direct applicability to APS AI governance or policy work.
- Week of 13 January 2025
-
AI Risk Profiles: A Standards Proposal for Pre-deployment AI Risk Disclosures
- A nine-category AI risk taxonomy proposes a standardised pre-deployment risk disclosure format for AI systems.
- The framework was applied to Claude, GPT APIs, Microsoft Copilot, GitHub Copilot, and Midjourney - tools common in APS contexts.
- Published in 2024 and spotlighted now via MIT AI Risk Repository; not a new standard but a peer-reviewed proposal.
-
MIT AI Risk Repository Selected for 2025 Paris Peace Forum AI Action Summit
- MIT AI Risk Repository was selected among 50 projects from 770 applications to present at the 2025 Paris AI Action Summit.
- The repository consolidates over 1,000 AI risks from 56 frameworks into a publicly accessible, categorised database with two taxonomies.
- The resource has over 100,000 visits and is already adopted by companies, governments, and researchers - a ready-made reference for APS risk work.
-
Sociotechnical harms of algorithmic systems: Scoping a taxonomy for harm reduction
- A 2023 AAAI/ACM paper taxonomises algorithmic harms into five categories and 20 subcategories.
- Categories span representational, allocative, quality-of-service, interpersonal, and social system harms across micro to macro levels.
- Spotlighted via MIT AI Risk Repository as a reference framework - useful for harm identification but not new or AU-specific.
- Week of 30 December 2024
-
Repository Updates: December 2024
- MIT AI Risk Repository expanded to 13 new frameworks in December 2024, now covering over 1000 AI risks.
- Added frameworks include Australian-authored contributions and a UK Government Office for Science frontier AI risks report.
- Repository is structured living knowledge infrastructure - quarterly updates planned through 2025 for researchers and policymakers.
-
Project Updates: December 2024
- MIT AI Risk Repository has reached 90,000 users since August 2024, with governments and companies using it globally.
- A new AI Risk Index project launches Q1 2025 to evaluate organisational responses to high-priority AI risks and identify gaps.
- Australian authors are among contributors to the Repository's frameworks, giving it some direct AU relevance.
-
Social Impacts of Artificial Intelligence and Mitigation Recommendations: An Exploratory Study
- A 2023 systematic review of 175 articles identifies nine categories of AI social impact, led by bias and discrimination.
- The MIT AI Risk Repository has catalogued this as one of its ten foundational risk frameworks for comparative reference.
- This is a 2021 conference paper spotlighted in early 2025 - findings are well-established rather than novel.
-
Managing the ethical and risk implications of rapid advances in artificial intelligence: A literature review
- MIT AI Risk Repository spotlights a 2016 literature review categorising AI ethical risks across three analytical axes.
- The framework uses PEST analysis to structure AI risk and proposes strategies like ethics committees and embedded ethics processes.
- The source paper is nearly a decade old; field has advanced significantly since, limiting direct applicability for current APS use.
-
NIST Researchers Meet with NHTSA Experts to Share Approaches to Assessment of Automated Vehicle System Performance
- NIST and NHTSA researchers met to share approaches to automated vehicle testing and performance assessment.
- Discussions focused on virtual and physical AV testing, sensor robustness, and scenario simulation - with potential collaboration to follow.
- Limited direct relevance to Australian federal agencies; AV standards work is at an early international stage.
-
NIST Hosts Second Stakeholder Workshop on Digital Twins
- NIST held its second stakeholder workshop on Digital Twins standards and interoperability in January 2025.
- Digital Twins are not AI systems; this item does not engage AI governance, models, or algorithmic decision-making.
- Limited direct relevance to Australian federal AI governance work - included for completeness only.
- Week of 23 December 2024
-
2024 in AI, with Nathan Benaich
- Annual AI roundup podcast covers model capabilities, frontier economics, robotics, compute, and regulation.
- Topics include global AI governance, industry consolidation, regulatory concerns, and 2025 predictions.
- Podcast format with limited citable depth - low priority for APS readers seeking actionable policy signal.
- Week of 16 December 2024
-
The Risks Associated with Artificial General Intelligence: A Systematic Review
- A 2023 systematic review identifies six AGI risk categories, from unsafe goals to existential risks.
- The MIT AI Risk Repository is cataloguing risk frameworks - a useful reference for APS risk taxonomy work.
- The reviewed literature relies heavily on philosophical discussion; limited peer-reviewed empirical risk modelling exists.
- Week of 9 December 2024
-
AI Policy and Governance Newsletter — December 2024
- Australia's Senate AI Select Committee released its final report recommending dedicated, principles-based AI legislation covering high-risk uses.
- The report's political fragmentation — dissenting Coalition report and Greens and Pocock additions — signals a difficult path for future AI legislation.
- Singapore's election deepfake ban and Canada's new AI Safety Institute provide international comparators directly relevant to Australian policy gaps.
- Week of 25 November 2024
-
Examining the differential risk from high-level artificial intelligence and the question of control
- MIT AI Risk Repository summarises a four-class risk taxonomy covering misuse, accident, structural, and agential AI risks.
- Expert survey data identifies monopolistic race dynamics, alignment failures, and power-seeking as highest-impact risks.
- This is a 2023 academic paper summary - useful background context but not new guidance for APS practitioners.
- Week of 18 November 2024
-
AI Generated Business: The Rise of AGI and the Rush to Find a Working Revenue Model
- AI Now Institute report argues OpenAI's business model emerged reactively, with AGI rhetoric serving as investor and marketing cover.
- The report contextualises how AI hype cycles—not sound economics—are driving hundreds of billions in investment into generative AI.
- Limited direct APS operational relevance; useful background for officials assessing vendor sustainability and AI market dynamics.
- Week of 11 November 2024
-
Shape, Symmetries, and Structure: The Changing Role of Mathematics in Machine Learning Research
- A Pacific Northwest National Laboratory researcher argues pure mathematics - topology, algebra, geometry - is increasingly relevant to ML research.
- The piece challenges the assumption that scaling existing models is sufficient for continued AI progress.
- Theoretical and academic in focus; limited direct relevance to APS AI governance or practice work.
- Week of 28 October 2024
-
Attorney-General's Department Annual Report 2023–24
- The Attorney-General's Department 2023–24 Annual Report covers performance and financials to 30 June 2024.
- No AI-specific content is indicated in the extracted text or item description.
- Limited direct relevance to APS AI governance - included for completeness only.
- Week of 21 October 2024
-
New AI Now Paper Highlights Risks of Commercial AI Used In Military Contexts
- Foundation models integrated into military intelligence, surveillance, and targeting systems pose significant and underappreciated national security risks.
- The paper argues CBRN proliferation concerns dominate policy debate while deployed AI targeting systems in Gaza present more immediate dangers.
- Recommendations to insulate military AI from commercial foundation models have indirect relevance to Australian Defence and intelligence AI procurement policy.
- Week of 14 October 2024
-
Redirecting Europe’s AI Industrial Policy
- AI Now Institute report challenges Europe's AI industrial policy focus on competitiveness and sovereignty over public benefit.
- Covers public procurement, cloud infrastructure, trade policy, and open AI as levers for redirecting AI investment.
- Limited direct APS relevance; offers comparative framing for Australian debates on sovereign AI and public-interest computing.
- Week of 7 October 2024
-
AI Policy and Governance Newsletter — October 2024
- Australia's Voluntary AI Safety Standard and mandatory guardrails discussion paper both use a 10-guardrail structure.
- Good Ancestors flags a key design flaw: one guardrail set applied across all AI types risks simultaneous over- and under-regulation.
- California's SB 1047 veto and UN Advisory Board AI governance proposals add international context for APS policy watchers.
-
Jacob Andreas: Language, Grounding, and World Models
- MIT professor Jacob Andreas discusses language grounding and world models in large language systems.
- Research focuses on computational foundations of language learning and systems that learn from human guidance.
- Extracted text is largely a bio stub - substantive podcast content is not captured in this item.
- Week of 30 September 2024
-
AI Now Associate Director Kate Brennan Testifies at the New York City Council Committee on Technology Hearing on the MyCity Portal
- AI Now testified against NYC's MyCity portal, citing corporate capture and citizen surveillance risks.
- Proposed digital wallets with spending tracking and 'incentive points' raise significant privacy and ethics concerns.
- US city-level case study; limited direct APS applicability but relevant to government-as-platform design debates.
-
A framework for ethical AI at the United Nations
- A 2021 UN-focused ethical AI framework identifies 13 risk categories from incompetence to lethal autonomous weapons.
- The MIT AI Risk Repository has catalogued this framework as one of six reference frameworks for AI risk taxonomy work.
- The paper is three years old and UN-scoped; APS practitioners likely have more current and jurisdiction-specific references available.
- Week of 23 September 2024
-
Mapping the ethics of generative AI: A comprehensive scoping review
- A scoping review identifies 378 normative issues across 19 ethical topic areas for generative AI systems.
- The taxonomy covers governance, fairness, safety, transparency, and hallucinations - all live APS concerns.
- The review notes literature imbalances, including disproportionate focus on risks and unsubstantiated scenarios.