Week of 30 September 2024
AI Now Institute testified against NYC's MyCity portal, citing corporate capture and citizen surveillance risks.
Key points
- The case illustrates risks when public AI infrastructure embeds vendor data advantages and opaque governance.
- Limited direct APS applicability, but the digital wallet and behavioural tracking scenario is a useful cautionary case study.
Week of 23 September 2024
A scoping review identifies 378 normative issues across 19 topic areas in generative AI ethics literature.
Key points
- The taxonomy covers areas directly relevant to APS AI governance: fairness, hallucinations, transparency, evaluation, and alignment.
- The MIT AI Risk Repository context makes this a useful reference for agencies building AI risk registers or ethics frameworks.
Podcast interview with journalist Evan Ratliff exploring his AI voice clone experiment for Shell Game.
Key points
- Discussion touches on ethical boundaries, societal impacts, and journalism's future with voice AI.
- Limited direct relevance to APS AI governance - included for context on public discourse around voice agents.
Week of 16 September 2024
MIT AI Risk Repository summarises a survey identifying seven core safety risks in generative language models.
Key points
- Risk categories include toxic content, hallucination, privacy leakage, and malicious use - directly relevant to APS AI governance frameworks.
- Survey is from 2023 (arXiv:2302.09270); useful as a taxonomy reference but not cutting-edge given rapid field evolution.
Week of 9 September 2024
A 2023 academic framework clusters AI ethics and responsibility issues into six groups via systematic literature review.
Key points
- The six clusters map closely to risk categories already recognised in Australian AI governance frameworks and agency guidance.
- This is a summary of an existing academic paper - useful context but not new primary guidance for APS practitioners.
Google DeepMind's Head of Human-AI Interaction Research discusses HCI, accessibility, and generative AI in depth.
Key points
- Topics include AGI definitions, anthropomorphisation, consent for generative clones, and bidirectional human-AI alignment.
- Academic podcast interview - thought-provoking but limited direct applicability to APS governance or policy work.
LLM benchmarks like MMLU and HumanEval may not reflect real user experience or collaborative utility.
Key points
- The piece argues current evaluation methods are non-interactive and ill-suited for human-AI collaboration models.
- Academic opinion piece from a Harvard PhD candidate - limited direct policy or APS operational relevance.
A Gradient Substack mini-update covers an international AI safety treaty and a technical codec development.
Key points
- The international AI safety treaty angle may be relevant to Australian AI governance and AISI positioning.
- Extracted text is paywalled - substantive content is not accessible for analysis.
The Gradient's issue 83 covers AI music streaming fraud, a new LLM search algorithm, and several AI news briefs.
Key points
- A legally binding Council of Europe AI treaty signed by the US, EU, UK, and others is briefly noted — Australia is not mentioned.
- Mixed-topic tech newsletter; no single item is developed in depth — low priority for focused APS reading.
APSC Commissioner finds 12 individuals breached the APS Code of Conduct 97 times in connection with Robodebt.
Key points
- Two former Secretaries, Campbell and Leon, named publicly - a rare exercise of discretion under s72A of the Public Service Act.
- Limited direct AI relevance; Robodebt's automated debt-raising mechanism is not the subject here - conduct and accountability are.
Week of 2 September 2024
A module-oriented LLM risk taxonomy covering 12 risks and 44 sub-categories across input, model, toolchain, and output layers.
Key points
- Included in the MIT AI Risk Repository, making it a reference point for agencies surveying structured AI risk frameworks.
- Primarily an academic arXiv paper summarised for practitioners - useful as background reading rather than actionable guidance.
Podcast interview with ARIA Programme Director Davidad Dalrymple covers provably safe AI and formal verification approaches.
Key points
- ARIA's Safeguarded AI Programme explores formal methods and Open Agency Architecture as technical safety pathways.
- Limited direct APS operational relevance; useful for practitioners tracking frontier AI safety research directions.
Week of 26 August 2024
TASRA classifies AI risks into six types based on accountability: who acts, whether unified, and whether deliberate.
Key points
- The taxonomy covers diffuse responsibility, unintended scale, willful indifference, criminal misuse, and state weaponisation.
- This is a 2023 academic preprint summarised in 2024 - useful reference material, not a new regulatory development.
A multi-topic AI newsletter covering copyright lawsuits, a novel misuse-prevention technique, and AI governance vignettes.
Key points
- The SOPHON research introduces a framework to prevent pre-trained models being fine-tuned for harmful or restricted tasks.
- Primarily US-focused content with limited direct APS relevance; useful as a broad AI landscape signal.
Week of 19 August 2024
A podcast interview with a Stanford cognitive scientist on human use of physical representations for sensemaking.
Key points
- AI is tangential - the lab uses AI methods but the focus is cognitive science and psychology.
- No direct relevance to APS AI governance, strategy, or policy work.
Week of 12 August 2024
A podcast interview with philosopher L.M. Sacasas on broad questions about technology and society.
Key points
- Covers philosophical themes - human embodiment, skills outsourcing, technological determinism - not AI governance.
- No direct relevance to Australian federal AI policy, governance, or APS practice.
Week of 5 August 2024
Good Ancestors' August 2024 newsletter covers deepfake legislation, US AI content bills, and regulatory sandboxes.
Key points
- Australia's Criminal Code Amendment (Deepfake Sexual Material) Bill 2024 has passed the Senate and is now in force.
- US legislative items are proposed only and not yet voted on; California's SB 1047 is flagged as a breaking development.
Week of 29 July 2024
AI Now Institute draws on FDA pharmaceutical regulation as a model for ex ante AI regulatory design.
Key points
- The report examines premarket scrutiny, regulatory functions, and industry capture risks - all live questions for Australian AI governance.
- Published mid-2024; the political climate the authors describe as hostile to premarket AI enforcement remains broadly unchanged.
Week of 8 July 2024
Good Ancestors' July 2024 newsletter covers three distinct AI policy developments across Australia and California.
Key points
- Two Australian items dominate: a joint federal-state-territory AI assurance framework and new deepfake criminalisation legislation.
- California's proposed $100M training-cost threshold for mandatory AI safety testing is the international item covered.
AI Now Institute testified to the US Senate that federal data privacy law is effectively AI regulation.
Key points
- Arguments centre on data minimisation, purpose limitation, and anti-monopoly checks on Big Tech AI development.
- US-focused advocacy testimony; limited direct applicability to Australian regulatory settings or APS practice.
Week of 1 July 2024
AI Now Institute launches a research program scrutinising Europe's emerging AI industrial policy and Big Tech dependencies.
Key points
- The piece warns that poorly designed industrial policy may entrench rather than challenge AI monopolies held by US and Chinese firms.
- Limited direct Australian policy relevance - useful context on global AI sovereignty debates Australia faces analogously.
Week of 24 June 2024
AI Now Institute argues military AI systems like Lavender and Gospel lack safety assurance, oversight, and accountability.
Key points
- The paper calls for safety engineering frameworks applied to military AI - directly relevant to defence AI governance debates.
- This is introductory framing for a future research series; substantive technical guidance is not yet published.
Week of 15 April 2024
AI Now Institute argues AI development is dominated by a handful of US and Chinese tech companies controlling core infrastructure.
Key points
- Speech advocates for public-interest AI industrial policy and rigorous EU AI Act implementation over market-led approaches.
- Limited direct APS relevance; a normative advocacy address to a European political audience, not a policy instrument or research finding.
Week of 11 March 2024
A small number of Big Tech firms control AI infrastructure, research incentives, and de facto standard-setting globally.
Key points
- Fragmented national regulation allows large AI firms to self-regulate and venue-shop, undermining effective governance.
- Item is a 2024 think-piece framing problems rather than offering operational guidance - signal value is contextual.