Research Assistant @GovAIOrg. Views my own.
London, England
Joined April 2011
- Tweets67.5K
- Following2.7K
- Followers7.3K
- Likes44.8K
Daniel Pryor retweeted
I distinctly remember lots of people saying this wouldn’t happen, because it wasn’t the intention. That’s not how laws work!
Daniel Pryor retweeted
Big new IFP report from @GeorgiaCAdamson, @KonstantinPilz, and me on the US-China AI competition.
We spent the last 6 months digging into how Chinese AI companies’ access to compute and distillation of American models translate into the current US-China capabilities gap.
The result is a 23k-word research report, including an interactive analytical model that visualizes changes in the US AI lead under different policy interventions.
Our findings:
1. The US AI lead over China is currently only around 6 months.
2. The US could roughly double this lead, to over a year, using authorities already available to the federal government.
3. On chips: China’s access to AI chips will roughly double over the next year, with chips housed in data centers overseas accounting for roughly half of China’s 2027 compute.
4. On distillation: the current ~16x US training compute advantage would normally yield a 13-month lead. But we estimate adversarial distillation halves this to around 6.5 months.
5. Increasing the lead would have some obvious benefits:
- Give the US broadly a greater cyberdefensive advantage against Chinese threat actors.
- Give US AI companies breathing room to spend more R&D resources on making their systems safer and more secure, without fear of being overtaken by less safety-conscious Chinese competitors.
6. Provide the US government more leverage in negotiations with China about the future of AI.
7. The two most important interventions are:
- For BIS: Create a new end-use control to restrict China’s remote access to training clusters abroad — China’s largest and fastest-growing source of frontier-scale compute.
- For DOJ/FTC: Create a narrow safe harbor to allow US companies to coordinate on countering Chinese adversarial distillation.
8. Both interventions (and all the others we discuss in the report) are things the federal government could do today.
9. To estimate the effects of each policy intervention, we model the US lead as a function of its compute advantage over China and China’s ability to adversarially distill the capabilities of the US frontier.
10. The model makes many simplifying assumptions, but provides a useful way to assess and rank the impact of different policy interventions at a high level.
Read the full report here: ifp.org/how-to-increase-amer…
Daniel Pryor retweeted
"The driver of Britain’s water shortages isn’t data centres, it is a 30 year failure to build the reservoirs we need to meet demand" open.substack.com/pub/samdum… @Sam_Dumitriu
Daniel Pryor retweeted
Exclusive:
The White House has reinstated the UK’s access to frontier AI models for safety testing after lobbying from ministers.
The US government reversed its decision to restrict access to the latest models from Anthropic, OpenAI and Google to stop the UK’s AI Safety Institute (AISI) from testing them before release.
Kanishka Narayan, the AI minister, has been involved in lobbying efforts to restore access.
Ministers feared that without the models, Britain’s attempt to become a world leader in AI regulation would be dead in the water.
telegraph.co.uk/news/2026/10…
Daniel Pryor retweeted
This is a slide from Terence Tao's 'Math 2.0' public lecture yesterday evening at Caltech. I am highly confident that, if the cure actually worked, almost any Stage 4 cancer patient would tell you that human comprehension of its mechanism is completely irrelevant to them.
Daniel Pryor retweeted
Published today in PNAS Nexus: Conversational AI increases political knowledge as effectively as self-directed internet search.
AISI’s Human Influence team finds that conversational AI's effects on political knowledge generalise across multiple topics, political perspectives, and model families, suggesting that the shift toward AI-assisted political information-seeking may not lead to increased public belief in political misinformation.
You can access the paper here: academic.oup.com/pnasnexus/a…
Daniel Pryor retweeted
This is, IMO, a pretty miserable calculus. I don’t want to have to beg for access to economically critical technologies. This is one reason to build a lot of compute now so that we have some leverage if it comes to that.
Daniel Pryor retweeted
Closer alignment with the EU, especially on security and defence, seems to now come with a major trade-off of making it harder to join whatever AI umbrella the US creates for aligned countries.
If we end up in a timeline where the US has much better models than the rest of the world, then having the option of adopting whatever measures it demands to get access to them (eg around national security, tech regulation, and defence) will be very important.
Daniel Pryor retweeted
Could automating AI R&D radically accelerate AI progress in an “intelligence explosion”?
Preliminary evidence suggests that it could.
In a new paper with authors across academia, civil society, and frontier AI companies (including @dawnsongtweets @merettm @jackclarkSF @Yoshua_Bengio @geoffreyhinton ), we assess the evidence & offer policy recommendations 🧵
Daniel Pryor retweeted
Third-party evaluators mostly test frontier AI models through an API. That misses the risks that come from how companies build and use AI internally.
Our new paper makes the case for embedded assessments. It was led by Jacob Charnock. The other co-authors are Sophie Williams, @zaheedkara, @Manderljung, Alejandro Tlaie Boria, @StephenLCasper, @AnkaReuel, and me.
🔗 Read the paper: governance.ai/research-paper…
Daniel Pryor retweeted
I am increasingly frustrated about the state of conversation about AI in the UK. It feels like we are in never ending cycles of deluded thinking. Under current trends, Britain will be defenceless against the transformative effects of AI and have no leverage over development at the frontier.
We are the most exposed economy to AI in the world. We have very little data centre capacity and few credible plans to expand – and our high energy costs mean we would struggle to even if we tried.
I am publishing a paper today on what the UK should do to build that leverage, let the economy capture value from AI and prepare the state to adapt, however AI develops.
The three strategies are as follows:
1. Build the infrastructure that gives us leverage. Fix the grid, fix planning, make Britain a place frontier AI can run.
2. Let the economy capture the value of AI. Reform the regulatory, labour and talent barriers that stop British firms adopting AI.
3. Build a state that can adapt to a changing world. Build a tax base that survives the labour market changing, public services that use AI to deliver better, and capacity to weather the upcoming geopolitical storms.
Two years ago, we published a similar report on AI - many recommendations of which ended up in the action plan. I remember feeling hopeful, but sadly the fundamental problems facing the country on AI have not shifted. I truly hope that this week of momentum culminates in action. There’s still a narrow path for the UK, but everyday that path is narrowing.
Daniel Pryor retweeted
Unless Britain sorts out its sky-high energy costs, it will abdicate whatever little remaining relevance it has to influence the future of AI.
(H/t @jujulemons)
Daniel Pryor retweeted
“A Transformative AI Strategy for Europe” is finally out. It argues that transformative AI could arrive before 2030, and that Europe isn’t ready.
The strategy was written by independent experts in academia and civil society, including @vestager, @Yoshua_Bengio, and @DAcemogluMIT. It makes 77 recommendations over 206 pages.
🔗 Read the strategy: transformative-ai.eu
🔗 Download the PDF: transformative-ai.eu/downloa…
Five objectives are flagged as immediate:
▶︎ Create a Member State Alliance for Supply Chain Security
Europe has no privileged access to frontier AI models. But it does host chokepoints in the AI supply chain (e.g. ASML’s lithography machines and Zeiss optics). The Netherlands, Germany, and France should found an alliance to turn those assets into leverage, mainly to protect against economic coercion.
▶︎ Make Europe’s institutions ready to act
The European Commission and Member State governments can’t compete for world-class frontier AI talent. The strategy asks for faster hiring and better pay, backed by ring-fenced funding. One estimate puts the cost of properly resourcing the AI Office’s supervisory work at around €50-60 million a year. The EU spends €278.8 million a year subsidizing banana growers in its outermost regions.
▶︎ Secure Europe’s share of global AI compute
Compute will be as important as energy, and Europe is a net importer of it. Data centers in Texas or Shenzhen can lock out European customers overnight. Europe hosts around 5% of global AI compute and should aim for 15% by 2030. Getting there means making data centers much faster to permit and connect.
▶︎ Ensure resilience to AI crises
Transformative AI would pose severe risks (e.g. from cyber attacks, biological weapons, and loss of control). Some of this is already happening. OpenAI, Anthropic, and the UK AI Security Institute have all lost control of AI agents in recent months. Europe needs hardened critical infrastructure and stockpiled crisis resources.
▶︎ Make Europe the global leader in AI assurance technology
Assurance technology verifies claims about AI models and the infrastructure they run on. Europe can’t show that a hosted model is secure, or that a foreign model is free of hidden backdoors (e.g. “secret loyalties”). So developers may be reluctant to host unreleased frontier models on EU soil. The same technology could verify whether countries keep to any future agreement to pace AI development.
Kudos to @MonikaSchnitzer and @privitera_ for initiating and steering the project. Glad I was able to contribute.
That was the most boring @F1 race I've watched in 20+ years of following the sport. Madring contract lasts until 2035 as well 🙄
Daniel Pryor retweeted
We're publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies.
These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve.
We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop.
Read the report: anthropic.com/threat-intelli…
Daniel Pryor retweeted
In 2026, AI policy is datacenter politics. You can protest them or tax them; you can attract them or send in the troops to shut them down. But in the next decade, the industry might leave the planet altogether.
Today, I take a first look at AI policy in the era of space compute: writing.antonleicht.me/p/esc…
Daniel Pryor retweeted
lots of middle powers want to bet on robotics, but they don't know how that relates to frontier AI: do they need frontier AI integration, or should they keep supply chains free from US presence?
this is a great piece on why frontier access might matter more than you'd think.
