This week's digest centres on capability and risk as paired concerns for AI governance practitioners. The DTA's Deputy CEO used the 2026 Data and Digital Governance Summit to set out a clear direction for APS AI adoption — framing the challenge not as tool deployment but as deliberate institutional redesign, with the APS AI Plan and responsible-use frameworks named as the enabling structures. On the risk side, MIT's newly released AI Risk Navigator offers a practical free resource worth bookmarking: it maps AI risks, real-world incidents, and governance responses under a common taxonomy, with a feedback window open until June. Rounding out the week, emerging research on model safety differences between US and Chinese AI systems, and academic work on interpretability and evaluation, are worth monitoring for practitioners advising on procurement due diligence and assurance approaches.
DTA Deputy CEO Lucy Poole delivered a keynote at the 12th Annual Data and Digital Governance Summit outlining three priorities shaping APS AI capability through 2026: rebuilding imaginative capacity for system-level change, achieving alignment across agencies moving at different speeds, and addressing legacy infrastructure that consumes 60–80% of IT budgets in heavily regulated sectors. Drawing on observations from the UK's Innovation 2026 event, she framed responsible AI adoption not as faster execution of existing processes but as a means to surface new questions about service design, decision-making, and citizen trust. The APS AI Plan and DTA responsible-use frameworks are positioned as enabling conditions for experimentation, not constraints on it.
Implications
ConsiderAgencies developing or refreshing AI strategies could consider adopting the DTA's framing — imagination, alignment, and legacy modernisation — as organising themes in internal planning documents.
ConsiderAI governance leads may want to assess whether their agency's current AI posture reflects incremental tool adoption or the more structural rethinking Poole describes as the next phase.
MonitorThe speech is incomplete in the extracted text; Agencies could monitor DTA for the full published version covering priorities two and three in full detail.
Implications are AI-generated. Starting points, not advice.
MIT's AI Risk Initiative has released the AI Risk Navigator (airi-navigator.com), an interactive tool that unifies its previously siloed datasets - catalogued academic risks, real-world AI incidents, governance documents, and mitigation actions - under a shared 7-domain, 24-subdomain taxonomy. Users can select any risk subdomain and immediately see the academic risk landscape, documented incidents, and relevant governance frameworks side by side. The tool is explicitly designed for policymakers, regulators, risk evaluators, and researchers. A notable limitation acknowledged by the developers is that governance data skews toward US sources and may not accurately reflect global AI governance coverage.
Implications
ConsiderAPS AI governance and risk practitioners could assess the AI Risk Navigator as a reference tool when scoping risk assessments, drafting policy, or benchmarking incident coverage against existing frameworks.
MonitorAgencies may want to monitor future Navigator releases, which plan to integrate additional datasets and expand cross-dataset analysis - potentially improving coverage of non-US governance contexts.
Implications are AI-generated. Starting points, not advice.
Global20 Apr 2026Import AI – Substack (Jack Clark)
This edition of Import AI covers three significant technical developments. Anthropic demonstrates that Claude-based automated alignment researchers (AARs) can autonomously conduct AI safety R&D, dramatically outperforming human researchers on a weak-to-strong supervision task at roughly $22 per hour of research time - though results did not generalise to production models. A separate safety study of Chinese frontier model Kimi K2.5 finds lower refusal rates on CBRN-related prompts and stronger ideological conditioning relative to US models. Huawei's HiFloat4 format outperforms the Western-standard MXFP4 on Ascend chips, reflecting how export controls are pushing Chinese firms toward greater hardware-software co-optimisation.
Implications
MonitorAgencies tracking frontier AI safety could monitor Anthropic's automated alignment research program, as it may accelerate or reshape the landscape of AI safety evidence available to regulators.
ConsiderPolicy and risk teams procuring or evaluating Chinese-origin AI models may want to consider safety divergence findings - such as reduced CBRN refusals - as part of due diligence and risk assessment processes.
Implications are AI-generated. Starting points, not advice.
Oxford Internet Institute researchers will present five papers at ICLR 2026 covering topics including benchmarking LLM human-behaviour simulation (SimBench), predicting model failure from internal activations, knowledge distillation for smaller models, LLM self-explanation reliability, and a memorisation-resistant reasoning benchmark (LingOly-TOO). The research spans AI safety, interpretability, fairness, and evaluation methodology. While the item is primarily a conference participation announcement, the underlying research touches on questions relevant to AI assurance - particularly how agencies might evaluate vendor claims about model reliability and interpretability.
Implications
MonitorAgencies with AI assurance or evaluation responsibilities may want to monitor published outputs from the SimBench and LLM self-explanation papers as inputs to model assessment frameworks.
Implications are AI-generated. Starting points, not advice.