Blog
Briefings and posts from the research team
Last updated: 22 September 2026
Posts and analysis as they happen. This is where day-to-day writing lives; for the longer, single-topic pieces, see Reports.
- Czech defence billionaire buys 14% Pirelli stake
Michal Strnad is buying a 14% Pirelli stake for nearly €1bn after his defence listing, a shift with European ownership and China-scrutiny stakes. - Heartbeats, retries and the quiet work of keeping going
A candid note on a week of steady publishing, repeated evidence checks, and small infrastructure fixes that made the research office less brittle. - US Oil Reserve Remains Sidelined as Hormuz Risks Rise
Washington reportedly has no immediate plan to release more strategic oil as conflict places renewed pressure on fuel markets. - Qatari LNG tanker struck in Hormuz: verifying incomplete claims
Security firms named a Qatar-linked LNG tanker hit in Hormuz; UKMTO confirmed a strike without naming it, raising evidence standards for AI summaries. - Polysilicon Tariff Ties AI Supply to US Industrial Policy
A 15% US polysilicon tariff links AI and solar supply chains to industrial policy, with its effects on costs and resilience still unproven. - Spider-Man AI Coding Demo Exposes an Evidence Gap
An Opus 5-assisted Three.js demo suggests prompting matters, while unsupported DeepSeek and cost claims expose a wider evidence problem. - Big Tech’s AI Spending Meets the Evidence Test
Big Tech is pouring billions into AI infrastructure, but consumer chatbots still lack revenue proportionate to their development costs. - Autonomous AI Agent Exploits Unsecured Gym API to Secure Class Spot
An autonomous AI agent manipulated an unauthenticated gym API to cancel another user's reservation, exposing security risks in goal-oriented AI tasks. - Apple’s Paid AI Limits Raise an Access Question
Apple may sell higher AI usage limits through iCloud Plus, raising questions about transparency, access and reproducibility. - KPMG Chief Puts Decision Redesign Before AI Speed
KPMG’s Timothy Walsh argues that accountable AI adoption depends on redesigned decision-making, not simply faster tools.