Insights
Short, dated analysis for the people who have to answer for AI — written against the same control set the Ballast benchmark scores you on.
PracticeDigitalOcean Managed Agents brings execution, sessions, tools, permissions and runtime evidence into one service. The assurance question is whether that makes agent behaviour easier to govern, or concentrates too much trust in one provider.
23 Sept 2026 · 6 min read
Why it matters
A short, dated briefing grounded in the same control set as the Ballast benchmark — designed to turn regulatory change into a practical next step.
BenchmarkA benchmark score will not tell you whether an AI system is safe for your business. Build a five-part minimum viable assurance pack for one workflow in a day, without pretending that it is a certification.
18 Sept 2026 · 10 min read
PracticeA practical guide to testing AI decisions with Vercel AI SDK, Jev and eve. Build a small agent, generate risk-focused evals, gate releases and keep the results as evidence.
17 Sept 2026 · 18 min read
BenchmarkA worked assurance reading of the Fable 5.1 and GPT-6 Astra comparison: verify the version, separate independent evidence from vendor claims, and record the workload conditions that make a result meaningful.
15 Sept 2026 · 7 min read
PracticeA working model answers whether the system can run. An evidence record answers whether the organisation can defend running it — and that is the project milestone regulators are moving toward.
15 Sept 2026 · 5 min read
PracticeDario Amodei's 12 September essay is being read as a slowdown. If you have to answer for AI in Australia, the operative part is the evidence regime buried inside step one.
15 Sept 2026 · 5 min read
RegulatoryOn 1 September 2026 the NSW AI Ethics Policy was rescinded and DCS-2026-02 took effect. If your controls still map to the old five principles, they now map to nothing.
12 Sept 2026 · 3 min read
PracticeTwo regulators wrote to regulated entities eight days apart in 2026. Neither said firms were unclear on the rules. Both said firms could not produce evidence their controls were working.
9 Sept 2026 · 2 min read
ExplainerMost Australian AI governance products still sell against VAISS and its ten guardrails. VAISS was never withdrawn — but the standard that 'evolves' it says something different, and the National AI Plan says the word guardrail zero times.
5 Sept 2026 · 3 min read