UK and US sign an AI pact — and fire a torpedo from a robot sub
Three guardrail stories landed on the same Tuesday, and only one of them came from a government interested in slowing anything down: a defence pact between two allies, a state cabinet in Illinois, and an independent testing institute for children's AI.
The UK and the US have launched a joint AI and autonomy partnership to protect critical national infrastructure, in the same week the two countries fired a torpedo from an uncrewed undersea vessel for the first time. The UK government said its Ministry of Defence Rapid AI Delivery Taskforce will work with the US Department of War's Chief Digital and Artificial Intelligence Office on AI and autonomy — systems meant to cover everything "from sea lanes to air space," detecting threats and deterring adversaries. The demonstration came before the paperwork: a heavyweight torpedo jointly developed by the US and Australia was launched from a 12-metre, Plymouth-built underwater drone named Excalibur in UK waters last week, an AUKUS milestone in which neither country had previously fired a torpedo from anything other than a crewed submarine.
The sequencing is the story. The partnership was announced as Prime Minister Andy Burnham arrived in New York for the UN General Assembly, where his address was expected to make the case for using AI for good, and where he held his first meeting with President Trump. It is the direct counter-position to the one taken from the same podium days earlier, which we covered here — Trump to the UN: no global AI rules, and AI is 'super intelligence'. Militaries are not waiting for the standards debate to finish: autonomy is already being tested in the water, and the agreements are following the hardware rather than leading it.
Illinois governor JB Pritzker signed an executive order creating an AI Cabinet that will consider tying data-center incentives and operating approvals to a company's safety, accountability and transparency standards. The unpaid group of outside experts and state agency representatives will produce policy proposals through 2027, assessing risks to residents and to the infrastructure they depend on — water systems, schools, information networks — while studying how to ease grid strain from data centers, whether state procurement can bake in AI accountability, and whether existing law is enough to hold AI companies accountable when they cause harm.
Illinois is not starting from zero. It already passed the country's first mandate for annual independent third-party audits of large frontier developers, with compliance due in 2028, and Pritzker has spent the month pressing Congress for national standards after an Anthropic researcher's resignation pushed AI safety onto the national agenda. What is new is the money lever — a state proposing to condition the incentives data centers want on standards the labs have so far written for themselves.
The former head of Google's trust and safety team has taken over Common Sense Media's Youth AI Safety Institute and wants AI products tested for children the way cars are crash-tested. Tom Siegel, who announced the move Tuesday, said the risks go past dependence and into "very, very severe scenarios of harm," naming suicide, psychosis and cognitive offloading, and accused the industry of repeating the mistakes of the social media era. His concrete ask: Google could let parents switch off AI Overviews and AI Mode in Search, as it already allows for sexually explicit content. Google said parents can block Search entirely and defended the features as a way for teenagers to learn; OpenAI said it combines account and behavioral signals with self-reported age to identify underage users.
What to watch: whether any state actually attaches conditions to a data-center approval, and whether the Youth AI Safety Institute's standards get adopted by anyone who is not already convinced.
Is a defence pact a guardrail or just a faster lane — and should militaries be writing the rules for autonomous systems at all? Tell us in the comments.
