700 Complaints, Four Platform Partners, No Court Case
Gulf AI: One Institution, Three Memoranda, One Number
The Agentic Civil Service Arrives, Rulebook Attached
Japan began moving 180,000 officials onto AI agents, Singapore named agent test metrics a contract could measure and Washington defined the incident categories for agents that act uninstructed — all in ten days. A comparison of what each regime requires of an agent deployer.
Read the story
The Briefing
Brazil's electoral court puts algorithmic reach on a six-hour clock, two AI Act duties fall due on 2 December, Pennsylvania gives advertisers eight weeks, Washington renames AI in its own paperwork, and six more moves from New York to Hobart. The week in ten items.
More stories
Beijing Puts a Jobs Test Inside AI Project Review
A State Council General Office opinion of 16 October 2026 requires an employment-risk assessment alongside project review for AI deployments in market-access sectors. Seoul, three days earlier, moved on AI training data instead. Two capitals, two theories of where to intervene.
Australia’s AI Build-Out Meets the Grid and the Councils
Two AI data-centre projects, in Tasmania and Victoria, stalled in the same week on electricity supply and local consent. Both decisions sit with sub-national bodies — and both land after an election or at year's end.
Geneva Gets a Standing Forum on Military AI
The UN's disarmament research institute has opened a permanent venue for talking about military AI, and the General Assembly has two draft resolutions on the table. Only one body can start a negotiation, and it meets in November.
Seven Days to Bid for India’s Sovereign GPUs
C-DAC opens a seven-day window on 22 October to empanel providers for up to 10,000 sovereign AI compute units. The entry bar is high, and New Delhi's promised data-centre policy still exists only as a ministerial remark.
What a Police AI Trial Measured, and What It Missed
Documents released under Right to Information show how Tasmania Police evaluated Axon's Draft One: a careful time study wrapped around a thin account of harm. Here is the evaluation design the next force should run.
The Ship That Wasn’t: Automation Bias Needs a Paper Trail
A chatbot's error reportedly came close to prompting a US intercept of a Chinese vessel. As AI tools reach millions of military personnel, high-stakes analysis needs provenance labels, verification gates and logs.
More opinionChina Turns Model Weights Into a Controlled Export
China's Announcement No. 48 treats installing a frontier Chinese model on foreign servers as a technology export. For governments building sovereign clouds on Chinese weights, the licence now matters as much as the chips.
Brazil Orders Meta’s AI to Treat Both Candidates Alike
Nine days before the run-off, Brazil's electoral court put Meta's ranking systems and its AI assistant under a parity duty, with a two-hour human-review clock, and wrote a test for when a WhatsApp channel belongs to a campaign.
Abu Dhabi Puts Its Government Platform Up for Licence
Abu Dhabi has created TAMM Global to license its AI-driven public-services platform to other governments, with an American city tipped as the first customer. The pitch rests on self-reported results and on conditions at home that no licence can transfer.
By region
Pennsylvania Makes Advertisers Confess Their AI
Act 99 of 2026 gives Pennsylvania advertisers eight weeks to disclose AI-made images, audio and video that could mislead consumers, with no knowledge defence. Colorado, meanwhile, is writing its AI rules through a regulator — and its only hearing is on Monday.
Explain the Algorithm: Brazil Sets a Six-Hour Clock
Three orders in forty-eight hours turned algorithmic reach into a reportable duty with a deadline attached. Brazil's electoral court now wants distribution data every six hours — and has platform staff sitting inside the courthouse.
Britain Prices Its AI Security State
HM Treasury has itemised roughly £456m for AI security, procurement and capability, to be confirmed at the Budget on 28 October 2026. The Commons Library sets out what the money does not buy: an AI act, or a regulator with powers.
Gulf AI: One Institution, Three Memoranda, One Number
In the week to 25 October 2026 the Gulf's AI news arrived as paperwork: a ratified Qatari decision creating a national AI centre, three memoranda from Oman's state visit to Britain, and one Saudi figure — 14MW — that can be checked on a date.
700 Complaints, Four Platform Partners, No Court Case
Nine days before South Africa votes in local elections, the Electoral Commission has logged more than 700 disinformation complaints, gazetted a code carrying R200,000 fines, and enlisted four platforms. Not one matter has reached the Electoral Court.
Pakistan Promises an Urdu Model and Sovereign Compute
Islamabad told a World Bank summit it is building an Urdu language model and buying sovereign compute. Neither announcement came with a tender, a budget line, a named builder or a date. Here is what each claim rests on.
Tasmania Makes AI Data Centres Pay Their Way
Hobart wants AI data centres to fund their host towns, publish water and power use every quarter and bring new renewables. A pulled float, a diesel dispute and a federal committee's blueprint show what the rules are up against.
Long Reads
The Decision Series. Four essays on how governments notice, weigh and act on what is happening around them.
The Decision Gap
Problems rarely arrive unannounced. The real failure of modern government is the distance between when a signal is available and when someone with authority acts on it, and that distance can be measured.
Beyond Dashboards
Governments have never counted more and decided less clearly. A review of 108 dashboard studies, and a stark gap between how officials and residents see public AI, suggest the chart of mentions is the wrong instrument.
Seeing the State in Motion
Government is not a single building but a moving system of tiers, departments and languages. One issue has many owners and many voices, and a state that reads only its dominant language sees only part of itself.
From Signal to Decision
Between noticing and acting lies the hardest stretch of government. On rumour, denial, silence and speed, and why the choice made at hour six should never be made on evidence fit only for hour sixty.
By desk
Pennsylvania Makes Advertisers Confess Their AI
Act 99 of 2026 gives Pennsylvania advertisers eight weeks to disclose AI-made images, audio and video that could mislead consumers, with no knowledge defence. Colorado, meanwhile, is writing its AI rules through a regulator — and its only hearing is on Monday.
Ottawa Says Data-Centre Power Is Not Its Problem
Canada's AI minister told senators on 22 October that regulating data-centre electricity is a provincial matter, days after announcing C$500m more for AI adoption. A dated map of who actually decides what, from Ottawa to Regina to Halifax.
Explain the Algorithm: Brazil Sets a Six-Hour Clock
Three orders in forty-eight hours turned algorithmic reach into a reportable duty with a deadline attached. Brazil's electoral court now wants distribution data every six hours — and has platform staff sitting inside the courthouse.
Gulf AI: One Institution, Three Memoranda, One Number
In the week to 25 October 2026 the Gulf's AI news arrived as paperwork: a ratified Qatari decision creating a national AI centre, three memoranda from Oman's state visit to Britain, and one Saudi figure — 14MW — that can be checked on a date.
What a Police AI Trial Measured, and What It Missed
Documents released under Right to Information show how Tasmania Police evaluated Axon's Draft One: a careful time study wrapped around a thin account of harm. Here is the evaluation design the next force should run.
