Week of 14 September 2026
A false AI-generated intelligence report nearly triggered a US military intercept of a Chinese vessel, almost starting armed conflict.
Key points
- An analyst used a chatbot to synthesise open-source and classified intelligence, producing an unverified assessment that entered operational workflows.
- The incident illustrates why high-consequence AI workflows require source provenance, confidence calibration, corroboration, and explicit human review gates.
EY survey of 202 US executives found 91% use agentic AI, but 49% have not updated governance frameworks for agent-specific risks.
Key points
- 26% of respondents using agentic AI could not detect unauthorised agents operating within their own organisations.
- Survey is self-reported and US-private-sector focused; findings are directionally useful but not independently audited.
Pentagon review found outdated intelligence and overreliance on Palantir's Maven targeting system contributed to a fatal school strike.
Key points
- AI did not autonomously select the target; human operators gave Maven excessive credit for detecting data quality problems.
- Illustrates automation bias and data provenance risks in high-consequence decision-support workflows - relevant to any APS ADM context.
Deloitte surveyed 25,000 UK workers; 63% used GenAI at work, 31% without employer knowledge.
Key points
- Shadow AI findings are directly analogous to risks APS agencies face as staff adopt consumer tools independently.
- Results are self-reported survey data, not a technical audit - governance gap is indicative, not precisely measured.
Microsoft opened a six-week public consultation on a draft Code of Conduct for its MAI models on 14 September 2026.
Key points
- The draft requires models to remain under human control, accept shutdown, and avoid self-directed goals - relevant to agentic AI procurement assessments.
- This is a consultation draft stating design intent, not a binding standard or evidence of implemented product controls.
New York's LOADinG Act, signed December 2024, mandates human review, impact assessments, and disclosure for state-agency automated decisions.
Key points
- The Act's workforce protections and procurement conditions offer a comparable model to consider alongside Australia's responsible AI policy for government.
- Scope is limited to New York state agencies affecting benefits, rights, or public employees - not a broad private-sector AI restriction.
US Defense Logistics Agency is piloting agentic AI 'digital employees' built on a foundation of roughly 185-190 bots.
Key points
- Governance measures include persona-based access controls, zero-trust security, data cleanup, and workforce supervision training.
- An early-stage pilot in a US defence logistics context - not yet a template for Australian agencies, but a useful comparator.
Camunda's vendor-commissioned survey of 6,000 employees found 40% of large organisations reported an AI compliance or governance issue in the past year.
Key points
- Process-related problems were linked to 84% of compliance incidents and 72% of AI initiative failures - pointing to workflow design as a governance risk factor.
- Survey is self-reported and vendor-commissioned, not regulator-confirmed; findings should be treated as indicative rather than authoritative.
Week of 7 September 2026
The AFP published its Artificial Intelligence Strategy 2026-28 on 2 September, covering investigation support, data analysis, and wellbeing.
Key points
- Five strategic goals span productivity, data-driven decisions, member wellbeing, workforce engagement, and countering criminal AI misuse.
- The strategy sets governance direction only; verifiable operational outcomes and disclosed risk assessments are the meaningful follow-on evidence.
UK NCSC warns that unapproved AI tool use creates data exposure, IP loss, and cyber security governance gaps.
Key points
- 71% of UK employees reportedly used employer-unapproved AI tools, suggesting shadow AI is a systemic problem, not an edge case.
- NCSC notes that restrictive controls without approved alternatives often push activity into unmonitored environments, worsening auditability.
Google's Q2 2026 threat intelligence identifies proprietary AI models, prompts, and credentials as active extortion and espionage targets.
Key points
- APS agencies holding AI assets - models, training data, deployment credentials - face equivalent exposure to private sector targets.
- Recommended controls mirror standard data governance: inventory, access separation, monitoring, and incident-response rehearsal.
Alan Turing Institute argues that workforce investment, not technology, is the primary driver of AI adoption success in national security.
Key points
- Defence and national security AI adoption hinges on human capability - a theme directly relevant to APS AI uplift strategies.
- Extracted text is truncated; full substance of the argument and any specific recommendations cannot be assessed from this item.
Microsoft, AFT, and UFT announced a legally enforceable AI privacy standard for US school districts on 9 September 2026.
Key points
- Terms prohibit using student or educator data to train AI, bar tracking, and require human oversight of AI decisions.
- Framework applies only when districts actively adopt terms in their Microsoft agreements - not automatic.
MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training released its final report on 25 August 2026.
Key points
- The report favours course- and discipline-specific AI expectations over a single institution-wide rule, treating it as a governance framework.
- Directly relevant to APS workforce training design; limited immediate federal policy implications for Australian agencies.
Week of 31 August 2026
A D.C. appellate court struck a Deutsche Bank brief after four AI-generated citations were confirmed nonexistent.
Key points
- The attorney used Google's generative-AI search tool without verifying citations - the court referred the matter for discipline.
- The ruling reinforces that verification controls, not just AI use policies, are required in high-consequence workflows.
Anthropic replaced mandatory 30-day server-side retention with customer-controlled Enterprise Frontier Safeguards after enterprise pushback.
Key points
- New controls allow automated safety monitoring without Anthropic human review - relevant to regulated-sector AI procurement assessments.
- Implementation details - encryption, key management, cloud boundaries, incident-response access - remain unspecified in published reporting.
The US federal government plans to pilot AI-led interviews via CodeSignal for its Tech Force recruitment program.
Key points
- OPM's August 27 memo establishes that AI outputs must not be the principal basis for consequential hiring decisions without independent human review.
- Australian context is absent; however, AI-assisted hiring governance mirrors emerging APS ADM and human-review obligations.
Dallas mounted AI cameras on 50 rubbish trucks to flag property code violations, scoring 21,000+ properties.
Key points
- The deployment lacks published accuracy metrics, false-positive rates, human-review procedures, or a resident appeal process.
- Australian agencies considering computer-vision enforcement tools face analogous governance gaps - this case illustrates them concretely.
Front evaluates AI agents on four independent axes—direction, coverage, faithfulness, safety—rejecting single-score evaluation.
Key points
- Synthetic test traffic drawn from documentation reliably inflates agent scores; real customer tickets are required for valid evaluation.
- Practical engineering detail from a private-sector deployment; no direct APS or Australian regulatory angle.
Scaling agentic AI beyond pilots requires connected data, back-end systems, and cohesive governance from the outset.
Key points
- Governance, privacy, security, and change management grow more critical as agents take on consequential work at scale.
- This is vendor-sponsored content from MIT Technology Review's custom arm - not independent editorial analysis.
AI inference workloads demand purpose-built infrastructure - memory, storage, and networking - not legacy enterprise IT.
Key points
- Performance per watt, cost efficiency, and scalability now matter as much as raw compute for AI deployments.
- This is vendor-adjacent infrastructure analysis; limited direct policy or governance relevance for APS practitioners.
A vendor-oriented case study on using AI to accelerate legacy system modernisation at Bupa health insurance.
Key points
- Key claims include four-times faster feature delivery, zero high-severity defects at launch, and AI-driven accessibility testing.
- Limited direct relevance to APS readers; private-sector healthcare context with no government or regulatory dimension.
Week of 24 August 2026
England and Wales police instructed to pause generative AI use in criminal justice work, including drafting court statements.
Key points
- A Copilot hallucination referencing a nonexistent football match illustrates the evidentiary risk in law enforcement AI use.
- The pause targets court-facing workflows specifically, not all police AI use - a meaningful governance distinction for APS agencies.
State Farm's outside counsel admitted AI-generated fake case citations appeared in Los Angeles court filings.
Key points
- The incident illustrates that fluent AI-generated prose does not guarantee accurate source grounding or citation validity.
- Direct APS regulatory consequence is low, but the pattern is directly relevant to any agency using AI in legal or policy drafting.
A community project maps four minimum viable agent incident-response controls to NIST CSF 2.0 functions, including a kill-switch contract.
Key points
- The specification addresses containment of agents acting with valid credentials - a gap conventional identity controls cannot close alone.
- No implementation results, adoption data, or independent validation are reported in available sources.