80000 Hours

Frontier AI companies have scaled their revenues, headcounts, and AI capabilities at a staggering rate in recent years. Since 2023, OpenAI and Anthropic have grown their staff by roughly 5–10x, revenue by over 30x, and AI models from simple chatbots to autonomous agents capable of expert-level cyberattacks and software engineering. If progress continues at this pace — or is accelerated by powerful AI systems automating AI research and development itself — the world will not be prepared.

When we talk about AI policy, usually we talk about sort of minimising the risk while capturing all the benefits. I think the default outcome for basically most countries in the world that don’t build their own frontier AI models is the exact opposite: they capture all the risk that comes from these models going through society, and yet they minimise all the benefits they could potentially have from AI happening.

We want to find creative and strategic and oftentimes totally experimental ways to build power for the cause of AI safety. This is obviously hard… because we’re working against the clock. And oftentimes we sink a lot of resources into things that we don’t know aren’t going to work until after the fact.

By · Last updated June 19th, 2026 · First published December 2022 · Summary In a nutshell: Many of the worst outcomes from advanced AI start with a security failure. If models can easily be stolen or manipulated, we’re all at greater risk from people who misuse them. If an AI system doesn’t have robust monitoring and control protocols, it might break out of its environment to pursue misaligned goals.

Intergalactic war is probably billions of years away — yet physics can already tell us how it ends. And strangely that conclusion is relevant to decisions people have to make today. In this video, Rob Wiblin walks through a fascinating analysis from researcher Beren Millidge that uses known physics — no wormholes or faster-than-light travel — to identify the only three weapons that could work at an intergalactic scale. We then unpack how to best defend against each.

What do professional advocates do? Advocacy is often misunderstood. While some people think advocacy means proving an argument, in practice it’s mostly about identifying shared interests and building coalitions of support around them. Making a compelling argument is often part of this, but it has to be complemented with strategically identifying decision makers and who they listen to, building trusted relationships, developing tacit knowledge about how decisions are made, etc.

Why work in AI policy within the US government? In a recent poll, 65% of Americans said that the government has done too little to regulate AI. But wanting more regulation and getting useful regulation are different things. In the 2010s, voters broadly supported federal action on the opioid epidemic, and Congress responded with major legislation in 2016 and 2018. These laws had some good provisions, but didn’t stop the crisis.

Why work on research on AI policy and strategy? In 2019, when researchers at the Center for Security and Emerging Technology (CSET) started working on AI policy research, it was a fairly niche topic. 2019 AIs couldn’t do much, and most policymakers weren’t thinking about the technology. But CSET started researching AI export controls, data protection, and cybersecurity. When AI development sped up over the next few years and the government needed expertise, they were ready.

By · Published June 8th, 2026 · We post over three thousand new jobs each year. Our top priority is to match those jobs with our readers. If you’re the right person for a job, we want to help you prove it. But even if you’re a strong candidate, you might struggle to break through.

Building an artifact like Gemini is very, very difficult. The main reason being you have to produce this one thing, this single set of model weights, deployed using a single serving stack. And it has to satisfy so many constraints… There’s probably 100 such things. And it is the case that if you make one change to the process with the intent of making one of these things better — say, safety — it will have random downstream knock-on effects on other constraints that you totally did not anticipate.