New releases

#229 – Marius Hobbhahn on the race to solve AI scheming before models go superhuman

Podcast by · Published December 3rd, 2025

We often worry about AI models “hallucinating” or making honest mistakes. But what happens when a model knows the truth, but decides to deceive you anyway to achieve a goal of its own? This isn’t sci-fi — it’s happening regularly in deployment today.

Marius Hobbhahn, CEO of the world’s top research organisation focused on AI deception (Apollo Research), has been collaborating with OpenAI to figure out when OpenAI’s reasoning models like o3 are inclined to “scheme” against the interests of users.

In a striking case uncovered by Apollo, when many AI models were told they would have capabilities removed if they performed too well on a test, they successfully “sandbagged” — intentionally answering questions incorrectly to appear less capable than they were, while also being careful not to perform so poorly it would arouse suspicion.

These models had somehow developed a preference to preserve their own capabilities, despite never being trained in that goal or assigned a task that called for it.

This doesn’t cause significant risk now, but as AI models become more general, superhuman in more areas, and are given more decision-making power, it could become outright dangerous.

In today’s episode, Marius details his recent collaboration with OpenAI to train o3 to follow principles like “never lie,” even when placed in “high-pressure” situations where it would otherwise make sense.

The good news: They reduced “covert rule violations” (scheming) by about 97%.

The bad news: In the remaining 3% of cases, the models sometimes became more sophisticated — making up new principles to justify their lying, or realising they were in a test environment and deciding to play along until the coast was clear.

Marius argues that while we can patch specific behaviours, we might be entering a “cat-and-mouse game” where models are becoming more situationally aware — that is, aware of when they’re being evaluated — faster than we are getting better at testing.

Even if models can’t tell they’re being tested, they can produce hundreds of pages of reasoning before giving answers and include strange internal dialects humans can’t make sense of, making it much harder to tell whether models are scheming or train them to stop.

Marius and host Rob Wiblin discuss:

  • Why models pretending to be dumb is a rational survival strategy
  • The Replit AI agent that deleted a production database and then lied about it
  • Why rewarding AIs for achieving outcomes might lead to them becoming better liars
  • The weird new language models are using in their internal chain-of-thought

This episode was recorded on September 19, 2025.

Video and audio editing: Dominic Armstrong, Milo McGuire, Luke Monsour, and Simon Monsour
Music: CORBIT
Camera operator: Mateo Villanueva Brandt
Coordination, transcripts, and web: Katy Moore

Continue reading →

Rob & Luisa chat kids, the fertility crash, and how the ‘50s invented parenting that makes us miserable

Podcast by and · Published November 25th, 2025

Global fertility rates aren’t just falling: the rate of decline is accelerating. From 2006 to 2016, fertility dropped gradually, but since 2016 the rate of decline has increased 4.5-fold. In many wealthy countries, fertility is now below 1.5. While we don’t notice it yet, in time that will mean the population halves every 60 years.

Rob Wiblin is already a parent and Luisa Rodriguez is about to be, which prompted the two hosts of the show to get together to chat about all things parenting — including why it is that far fewer people want to join them raising kids than did in the past.

While “kids are too expensive” is the most common explanation, Rob argues that money can’t be the main driver of the change: richer people don’t have many more children now, and we see fertility rates crashing even in countries where people are getting much richer.

Instead, Rob points to a massive rise in the opportunity cost of time, increasing expectations parents have of themselves, and a global collapse in socialising and coupling up. In the EU, the rate of people aged 25–35 in relationships has dropped by 20% since 1990, which he thinks will “mechanically reduce the number of children.” The overall picture is a big shift in priorities: in the US in 1993, 61% of young people said parenting was an important part of a flourishing life for them, vs just 26% today.

That leads Rob and Luisa to discuss what they might do to make the burden of parenting more manageable and attractive to people, including themselves.

In this non-typical episode, we take a break from the usual heavy topics to discuss the personal side of bringing new humans into the world, including:

  • Rob’s updated list of suggested purchases for new parents
  • How parents could try to feel comfortable doing less
  • How beliefs about childhood play have changed so radically
  • What matters and doesn’t in childhood safety
  • Why the decline in fertility might be impractical to reverse
  • Whether we should care about a population crash in a world of AI automation

This episode was recorded on September 12, 2025.

Video and audio editing: Dominic Armstrong, Milo McGuire, Luke Monsour, and Simon Monsour
Music: CORBIT
Camera operator: Jeremy Chevillotte
Coordination, transcripts, and web: Katy Moore

Continue reading →

#228 – Eileen Yam on how we’re completely out of touch with what the public thinks about AI

Podcast by · Published November 20th, 2025

If you work in AI, you probably think it’s going to boost productivity, create wealth, advance science, and improve your life. If you’re a member of the American public, you probably strongly disagree.

In three major reports released over the last year, the Pew Research Center surveyed over 5,000 US adults and 1,000 AI experts. They found that the general public holds many beliefs about AI that are virtually nonexistent in Silicon Valley, and that the tech industry’s pitch about the likely benefits of their work has thus far failed to convince many people at all. AI is, in fact, a rare topic that mostly unites Americans — regardless of politics, race, age, or gender.

Today’s guest, Eileen Yam, director of science and society research at Pew, walks us through some of the eye-watering gaps in perception:

  • Jobs: 73% of AI experts see a positive impact on how people do their jobs. Only 23% of the public agrees.
  • Productivity: 74% of experts say AI is very likely to make humans more productive. Just 17% of the public agrees.
  • Personal benefit: 76% of experts expect AI to benefit them personally. Only 24% of the public expects the same (while 43% expect it to harm them).
  • Happiness: 22% of experts think AI is very likely to make humans happier, which is already surprisingly low — but a mere 6% of the public expects the same.

For the experts building these systems, the vision is one of human empowerment and efficiency. But outside the Silicon Valley bubble, the mood is more one of anxiety — not only about Terminator scenarios, but about AI denying their children “curiosity, problem-solving skills, critical thinking skills and creativity,” while they themselves are replaced and devalued:

  • 53% of Americans say AI will worsen people’s ability to think creatively.
  • 50% believe it will hurt our ability to form meaningful relationships.
  • 38% think it will worsen our ability to solve problems.

Open-ended responses to the surveys reveal a poignant fear: that by offloading cognitive work to algorithms we are changing childhood to a point we no longer know what adults will result. As one teacher quoted in the study noted, we risk raising a generation that relies on AI so much it never “grows its own curiosity, problem-solving skills, critical thinking skills and creativity.”

If the people building the future are this out of sync with the people living in it, the impending “techlash” might be more severe than industry anticipates.

In this episode, Eileen and host Rob Wiblin break down the data on where these groups disagree, where they actually align (nobody trusts the government or companies to regulate this), and why the “digital natives” might actually be the most worried of all.

This episode was recorded on September 25, 2025.

Video and audio editing: Dominic Armstrong, Milo McGuire, Luke Monsour, and Simon Monsour
Music: CORBIT
Coordination, transcripts, and web: Katy Moore

Continue reading →

The US AI policy landscape: where to have the biggest impact

Article · Published November 17th, 2025

Aerial view of Washington, DC at dusk showing the illuminated US Capitol building in the centre, with the Washington Monument visible in the background to the left. The city lights glow against an orange and purple sunset sky.

The US government may be the single most important actor for shaping how AI develops. If you want to improve the trajectory of AI and reduce catastrophic risks, you could have an outsized impact by working on US policy.

But the US policy ecosystem is huge and confusing. And policy shaping AI is made by specific people in specific places — so where you work matters enormously.

This guide aims to help you think about where specifically to work in US AI policy so you can actually make a large impact.

In Part 1, we cover five heuristics for finding the most impactful places to work. In Part 2, we cover five policy institutions that we’d guess are the most impactful for AI and name specific places in each that are especially promising.

Jump to our top recommended institutions

Part 1: How to find the most impactful places to work on AI policy

It’s hard to predict precisely when and where key AI policy decisions will happen, but you can position yourself for greater impact. The following five heuristics can help you judge where you could have the best shot at positively shaping the trajectory of advanced AI.

Prioritise building career capital

Early in your policy career, avoid tunnel vision on AI policy roles. Many entry-level positions worth considering won’t focus on AI. What usually matters more is building career capital — knowledge,

Continue reading →

OpenAI: The nonprofit refuses to die (with Tyler Whitmer)

Podcast by · Published November 11th, 2025

Last December, the OpenAI business put forward a plan to completely sideline its nonprofit board. But two state attorneys general have now blocked that effort and kept that board very much alive and kicking.

The for-profit’s trouble was that the entire operation was founded on the premise of — and legally pledged to — the purpose of ensuring that “artificial general intelligence benefits all of humanity.” So to get its restructure past regulators, the business entity has had to agree to 20 serious requirements designed to ensure it continues to serve that goal.

Attorney Tyler Whitmer, as part of his work with Legal Advocates for Safe Science and Technology, has been a vocal critic of OpenAI’s original restructure plan. In today’s conversation, he lays out all the changes and whether they will ultimately matter:

Not For Private Gain chart

After months of public pressure and scrutiny from the attorneys general (AGs) of California and Delaware, the December proposal itself was sidelined — and what replaced it is far more complex and goes a fair way towards protecting the original mission:

  • The nonprofit’s charitable purpose — “ensure that artificial general intelligence benefits all of humanity” — now legally controls all safety and security decisions at the company. The four people appointed to the new Safety and Security Committee can block model releases worth tens of billions.
  • The AGs retain ongoing oversight, meeting quarterly with staff and requiring advance notice of any changes that might undermine their authority.
  • OpenAI’s original charter, including the remarkable “stop and assist” commitment, remains binding.

But significant concessions were made. The nonprofit lost exclusive control of AGI once developed — Microsoft can commercialise it through 2032. And transforming from complete control to this hybrid model represents, as Tyler puts it, “a bad deal compared to what OpenAI should have been.”

The real question now: will the Safety and Security Committee use its powers? It currently has four part-time volunteer members and no permanent staff, yet they’re expected to oversee a company racing to build AGI while managing commercial pressures in the hundreds of billions.

Tyler calls on OpenAI to prove they’re serious about following the agreement:

  • Hire management for the SSC.
  • Add more independent directors with AI safety expertise.
  • Maximise transparency about mission compliance.

There’s a real opportunity for this to go well. A lot … depends on the boards, so I really hope that they … step into this role … and do a great job. … I will hope for the best and prepare for the worst, and stay vigilant throughout.

Host Rob Wiblin and Tyler discuss all that and more in today’s episode.

This episode was recorded on November 4, 2025.

Video editing: Milo McGuire, Dominic Armstrong, and Simon Monsour
Audio engineering: Milo McGuire, Simon Monsour, and Dominic Armstrong
Music: CORBIT
Coordination, transcriptions, and web: Katy Moore

Continue reading →

#227 – Helen Toner on the geopolitics of AI in China and the Middle East

Podcast by · Published November 5th, 2025

With the US racing to develop AGI and superintelligence ahead of China, you might expect the two countries to be negotiating how they’ll deploy AI, including in the military, without coming to blows. But according to Helen Toner, director of the Center for Security and Emerging Technology in DC, “the US and Chinese governments are barely talking at all.”

In her role as a founder, and now leader, of DC’s top think tank focused on the geopolitical and military implications of AI, Helen has been closely tracking the US’s AI diplomacy since 2019.

“Over the last couple of years there have been some direct [US–China] talks on some small number of issues, but they’ve also often been completely suspended.” China knows the US wants to talk more, so “that becomes a bargaining chip for China to say, ‘We don’t want to talk to you. We’re not going to do these military-to-military talks about extremely sensitive, important issues, because we’re mad.'”

Helen isn’t sure the groundwork exists for productive dialogue in any case. “At the government level, [there’s] very little agreement” on what AGI is, whether it’s possible soon, whether it poses major risks. Without shared understanding of the problem, negotiating solutions is very difficult.

Another issue is that so far the Chinese Communist Party doesn’t seem especially “AGI-pilled.” While a few Chinese companies like DeepSeek are betting on scaling, she sees little evidence Chinese leadership shares Silicon Valley’s conviction that AGI will arrive any minute now, and export controls have made it very difficult for them to access compute to match US competitors.

When DeepSeek released R1 just three months after OpenAI’s o1, observers declared the US–China gap on AI had all but disappeared. But Helen notes OpenAI has since scaled to o3 and o4, with nothing to match on the Chinese side. “We’re now at something like a nine-month gap, and that might be longer.”

To find a properly AGI-pilled autocracy, we might need to look at nominal US allies. The US has approved massive data centres in the UAE and Saudi Arabia with “hundreds of thousands of next-generation Nvidia chips” — delivering colossal levels of computing power.

When OpenAI announced this deal with the UAE, they celebrated that it was “rooted in democratic values,” and would advance “democratic AI rails” and provide “a clear alternative to authoritarian versions of AI.”

But the UAE scores 18 out of 100 on Freedom House’s democracy index. “This is really not a country that respects rule of law,” Helen observes. Political parties are banned, elections are fake, dissidents are persecuted.

If AI access really determines future national power, handing world-class supercomputers to Gulf autocracies seems pretty questionable. The justification is typically that “if we don’t sell it, China will” — a transparently false claim, given severe Chinese production constraints. It also raises eyebrows that Gulf countries conduct joint military exercises with China and their rulers have “very tight personal and commercial relationships with Chinese political leaders and business leaders.”

In today’s episode, host Rob Wiblin and Helen discuss the above, plus:

  • Ways China exaggerates its chip production for strategic gain
  • The confusing and conflicting goals in the US’s AI policy towards China
  • Whether it matters that China could steal frontier AI models trained in the US
  • Whether Congress is starting to take superintelligence seriously this year
  • Why she rejects ‘non-proliferation’ as a model for AI
  • Plenty more.

CSET is hiring! Check out its careers page for current roles.

This episode was recorded on September 25, 2025.

Video editing: Luke Monsour and Simon Monsour
Audio engineering: Milo McGuire, Simon Monsour, and Dominic Armstrong
Music: CORBIT
Coordination, transcriptions, and web: Katy Moore

Continue reading →

Our top tips for successful networking

Blog post by · Published November 5th, 2025

Everyone talks about the importance of networking for a successful career. And they’re right — the people you connect with will shape your habits, the ideas you’re exposed to, and your job opportunities.

But how do you actually network well?

I asked this question to our career advisors, who have helped thousands of people break into high-impact roles. Here’s what they recommended.

How to network

The basics are simple: find people who can help you learn or move forward with your career, or who you can help. Increase your opportunities to connect with them, and try to build genuine relationships with the people you meet.

Find the right people

  • Attend conferences, courses, and social events in your professional community, or one you’d like to be a part of. There are many opportunities to meet like-minded people in person and virtually — we have lots of recommendations to get you started on our community page.
  • Get involved in online discourse. People often use social media, forums, and other online platforms to connect with others and discuss ideas. Whatever your interests are, there’s probably an online community out there for you! Our advisors sometimes recommend being active on X/Twitter, especially if you’re interested in AI policy.
  • Try visiting a hub. There are some locations (like the San Francisco Bay Area, Washington DC, London, and Oxford) with a large number of people interested in existential risks and effective altruism.

Continue reading →

Open positions to grow our podcast team

Blog post by · Published October 31st, 2025

Why this matters

The 80,000 Hours Podcast aims to help the world safely navigate the transition to a flourishing future with artificial general intelligence. We focus on influencing the decisions that matter most — helping people identify high-leverage career moves and important actions they can take in their current positions.

Our episodes reach tens or hundreds of thousands of people, and we regularly hear from employees in government and at leading AI companies who find them useful. We’re also growing fast: in 2025, our listenership was around 60% higher than the same period last year. In order to further improve the usefulness and reach of our episodes, we need to increase the team’s capacity.

Right now, in addition to recording important conversations, our hosts handle much of the crucial work that surrounds each episode:

  • Identifying which conversations would actually change important decisions
  • Finding guests who can deliver those insights
  • Shaping episodes to land the key arguments effectively
  • Ensuring the right decision-makers encounter the content at the right moment

We’re looking to hire two people to the team who can take on some of this work, allowing Rob and Luisa to focus more of their time on making great episodes.

The window is short. If we’re right about AI timelines, the decisions that shape humanity’s long-run trajectory are happening now, not in 20 years. We’re a small team trying to do something unreasonably ambitious and we’re looking for the right people to help us scale up.

Continue reading →

#226 – Holden Karnofsky on dozens of amazing opportunities to make AI safer — and all his AGI takes

Podcast by · Published October 30th, 2025

For years, working on AI safety usually meant theorising about the ‘alignment problem’ or trying to convince other people to give a damn. If you could find any way to help, the work was frustrating and low feedback.

According to Anthropic’s Holden Karnofsky, this situation has now reversed completely.

There are now large amounts of useful, concrete, shovel-ready projects with clear goals and deliverables. Holden thinks people haven’t appreciated the scale of the shift, and wants everyone to see the large range of ‘well-scoped object-level work’ they could personally help with, in both technical and non-technical areas.

In today’s interview, Holden — previously cofounder and CEO of Open Philanthropy (now Coefficient Giving) — lists 39 projects he’s excited to see happening, including:

  • Training deceptive AI models to study deception and how to detect it
  • Developing classifiers to block jailbreaking
  • Implementing security measures to stop ‘backdoors’ or ‘secret loyalties’ from being added to models in training
  • Developing policies on model welfare, AI-human relationships, and what instructions to give models
  • Training AIs to work as alignment researchers

And that’s all just stuff he’s happened to observe directly, which is probably only a small fraction of the options available.

All this low-hanging fruit is one factor behind his decision to join Anthropic this year. That said, his wife was also a cofounder and president of the company, giving him a big financial stake in its success — and making it impossible for him to be seen as independent no matter where he worked.

Holden makes a case that, for many people, working at an AI company like Anthropic will be the best way to steer AGI in a positive direction. He notes there are “ways that you can reduce AI risk that you can only do if you’re a competitive frontier AI company.” At the same time, he believes external groups have their own advantages and can be equally impactful.

Outside critics worry that Anthropic’s efforts to stay at that frontier encourage competitive racing towards AGI — significantly or entirely offsetting any useful research they do. Holden thinks this seriously misunderstands the strategic situation we’re in.

“I work at an AI company, and a lot of people think that’s just inherently unethical,” he says. “They’re imagining [that] everyone wishes they could go slowly, but they’re going fast so they can beat everyone else. […] But I emphatically think this is not what’s going on in AI.”

The reality, in Holden’s view:

I think there’s too many players in AI who […] don’t want to slow down. They don’t believe in the risks. Maybe they don’t even care about the risks. […] If Anthropic were to say, “We’re out, we’re going to slow down,” they would say, ‘This is awesome! Now we have a better chance of winning, and this is even good for our recruiting” — because they have a better chance of getting people who want to be on the frontier and want to win.

Holden believes a frontier AI company can reduce risk by:

  • Developing cheap, practical safety measures other companies might adopt
  • Prototyping policies regulators could mandate
  • Gathering crucial data about what advanced AI can actually do

Host Rob Wiblin and Holden discuss the case for and against those strategies, and much more, in today’s episode.

This episode was recorded on July 25 and 28, 2025.

Video editing: Simon Monsour, Luke Monsour, Dominic Armstrong, and Milo McGuire
Audio engineering: Milo McGuire, Simon Monsour, and Dominic Armstrong
Music: CORBIT
Coordination, transcriptions, and web: Katy Moore

Continue reading →

Extreme power concentration

Problem profile by · Last updated October 28th, 2025 · First published April 2025

Advanced AI technology may enable its creators, or others who control it, to attempt and achieve unprecedented societal power grabs. Under certain circumstances, they could use these systems to take control of whole economies, militaries, and governments.

This kind of power grab from a single person or small group would pose a major threat to the rest of humanity.

Continue reading →

#225 – Daniel Kokotajlo on what a hyperspeed robot economy might look like

Podcast by · Published October 20th, 2025

When Daniel Kokotajlo talks to security experts at major AI labs, they tell him something chilling: “Of course we’re probably penetrated by the CCP already, and if they really wanted something, they could take it.”

This isn’t paranoid speculation. It’s the working assumption of people whose job is to protect frontier AI models worth billions of dollars. And they’re not even trying that hard to stop it — because the security measures that might actually work would slow them down in the race against competitors.

Daniel is the founder of the AI Futures Project and author of AI 2027, a detailed scenario showing how we might get from today’s AI systems to superintelligence by the end of the decade. Over a million people read it in the first few weeks, including US Vice President JD Vance. When Daniel talks to researchers at Anthropic, OpenAI, and DeepMind, they tell him the scenario feels less wild to them than to the general public — because many of them expect something like this to happen.

Daniel’s median timeline? 2029. But he’s genuinely uncertain, putting 10–20% probability on AI progress hitting a long plateau.

When he first published AI 2027, his median forecast for when superintelligence would arrive was 2027, rather than 2029. So what shifted his timelines recently? Partly a fascinating study from METR showing that AI coding assistants might actually be making experienced programmers slower — even though the programmers themselves think they’re being sped up. The study suggests a systematic bias toward overestimating AI effectiveness — which, ironically, is good news for timelines, because it means we have more breathing room than the hype suggests.

But Daniel is also closely tracking another METR result: AI systems can now reliably complete coding tasks that take humans about an hour. That capability has been doubling every six months in a remarkably straight line. Extrapolate a couple more years and you get systems completing month-long tasks. At that point, Daniel thinks we’re probably looking at genuine AI research automation — which could cause the whole process to accelerate dramatically.

At some point, superintelligent AI will be limited by its inability to directly affect the physical world. That’s when Daniel thinks superintelligent systems will pour resources into robotics, creating a robot economy in months.

Daniel paints a vivid picture: imagine transforming all car factories (which have similar components to robots) into robot production factories — much like historical wartime efforts to redirect production of domestic goods to military goods. Then imagine the frontier robots of today hooked up to a data centre running superintelligences controlling the robots’ movements to weld, screw, and build. Or an intermediate step might even be unskilled human workers coached through construction tasks by superintelligences via their phones.

There’s no reason that an effort like this isn’t possible in principle. And there would be enormous pressure to go this direction: whoever builds a superintelligence-powered robot economy first will get unheard-of economic and military advantages.

From there, Daniel expects the default trajectory to lead to AI takeover and human extinction — not because superintelligent AI will hate humans, but because it can better pursue its goals without us.

But Daniel has a better future in mind — one he puts roughly 25–30% odds that humanity will achieve. This future involves international coordination and hardware verification systems to enforce AI development agreements, plus democratic processes for deciding what values superintelligent AIs should have — because in a world with just a handful of superintelligent AI systems, those few minds will effectively control everything: the robot armies, the information people see, the shape of civilisation itself.

Right now, nobody knows how to specify what values those minds will have. We haven’t solved alignment. And we might only have a few more years to figure it out.

Daniel and host Luisa Rodriguez dive deep into these stakes in today’s interview.

This episode was recorded on September 9, 2025.

Audio engineering: Milo McGuire, Simon Monsour, and Dominic Armstrong
Music: CORBIT
Coordination, transcriptions, and web: Katy Moore

Continue reading →

If we can’t control MechaHitler, how will we steer AGI?

Blog post by and · Published October 9th, 2025

It’s Chana and Aric again.

Firstly, we cannot believe the response to our last video. Thank you all so much for creating an initial swell of views and support, for your feedback and comments and subscriptions. We’re at 186,000 subscribers, and 6 million views. Truly, truly, incredible; we’re so delighted to be doing this with you.

Also, um, feel free to do that all again? Because AI in Context has a second video for you.

There are two types of people in the world:

  1. People who are understandably very confused about why the word “MechaHitler” appears in the title of this sober, thoughtful, serious blog post.
  2. People grimacing because they know why “MechaHitler” appears in the title of this sober, thoughtful, serious blog post.

If you’re in category 1, here’s the summary:

You might have heard of Elon Musk’s AI model, Grok, which can interact with users and post directly to Twitter. Earlier this year, it suddenly turned from being a fairly neutral commentator on events to a sexually-harassing, Nazi-minded troll calling itself ‘MechaHitler.’

Our new video is about that incident, getting into the wild details of the MechaHitler story that even the terminally online may have missed.

But it’s also about putting this news story in context: how Elon Musk and his AI company xAI function, where the broader AI industry is going, and what this incident reveals about our readiness for more powerful AI systems.

Continue reading →

#224 – Andrew Snyder-Beattie on the low-tech plan to patch humanity’s greatest weakness

Podcast by · Published October 2nd, 2025

Conventional wisdom is that safeguarding humanity from the worst biological risks — microbes optimised to kill as many as possible — is difficult bordering on impossible, making bioweapons humanity’s single greatest vulnerability. Andrew Snyder-Beattie thinks conventional wisdom could be wrong.

Andrew’s job at Open Philanthropy (now Coefficient Giving) is to spend hundreds of millions of dollars to protect as much of humanity as possible in the worst-case scenarios — those with fatality rates near 100% and the collapse of technological civilisation a live possibility.

As Andrew lays out, there are several ways this could happen, including:

  • A national bioweapons programme gone wrong (most notably Russia or North Korea’s)
  • AI advances making it easier for terrorists or a rogue AI to release highly engineered pathogens
  • Mirror bacteria that can evade the immune systems of not only humans, but many animals and potentially plants as well

Most efforts to combat these extreme biorisks have focused on either prevention or new high-tech countermeasures. But prevention may well fail, and high-tech approaches can’t scale to protect billions when, with no sane person willing to leave their home, we’re just weeks from economic collapse.

So Andrew and his biosecurity research team at Open Philanthropy have been seeking an alternative approach. They’re now proposing a four-stage plan using simple technology that could save most people, and is cheap enough it can be prepared without government support.

Andrew is hiring for a range of roles to make it happen — from manufacturing and logistics experts to global health specialists to policymakers and other ambitious entrepreneurs.

The approach exploits tiny organisms having no way to penetrate physical barriers or shield themselves from UV, heat, or chemical poisons.

We now know how to make highly effective ‘elastomeric’ face masks that cost $10, can sit in storage for 20 years, and can be used for six months straight without changing the filter. Any rich country could trivially stockpile enough to cover all essential workers.

People can’t wear masks 24/7, but fortunately propylene glycol — already found in vapes and smoke machines — is astonishingly good at killing microbes in the air. And, being a common chemical input, industry already produces enough of the stuff to cover every indoor space we need at all times.

Add to this the wastewater monitoring and metagenomic sequencing that will detect the most dangerous pathogens before they have a chance to wreak havoc, and we might just buy ourselves enough time to develop the cure we’ll need to come out alive.

Has everyone been wrong, and biology is actually defence dominant rather than offence dominant? Is this plan crazy — or so crazy it just might work?

That’s what host Rob Wiblin and Andrew Snyder-Beattie explore in this in-depth conversation.

This episode was recorded on August 12, 2025

Video editing: Simon Monsour and Luke Monsour
Audio engineering: Milo McGuire, Simon Monsour, and Dominic Armstrong
Music: CORBIT
Camera operator: Jake Morris
Coordination, transcriptions, and web: Katy Moore

Continue reading →

Survey results: what AI safety orgs want in a hire

Blog post by · Published September 23rd, 2025

Hi everyone!

A few months ago, Benjamin Todd and I surveyed top figures in AI safety about the hiring needs of their organisations, including:

  • What sort of people they’re hiring in the coming years
  • How people can skill up to get a job
  • The hardest qualities to find in applicants

In total, we heard from 38 people running top organisations working on making AGI go well for humanity. Some very interesting suggestions emerged that I’d like to share with job hunters:

What are the best steps for upskilling? A few programmes were mentioned multiple times. In order:

Several respondents said that if you’re talented or have 1–3 years of experience building a relevant skill set, a BlueDot course or fellowship can be sufficient AI context for getting hired into positions (excluding roles in technical AI research).

Startups were also recommended multiple times as a good place to gain operations experience, engineering experience, and “experience at moving fast.”

What about policy experience? If you’re trying to enter an AI policy role, general policy experience is important,

Continue reading →

#223 – Neel Nanda on leading a Google DeepMind team at 26 – and advice if you want to work at an AI company (part 2)

Podcast by · Published September 15th, 2025

At 26, Neel Nanda leads an AI safety team at Google DeepMind, has published dozens of influential papers, and mentored 50 junior researchers — seven of whom now work at major AI companies. His secret? “It’s mostly luck,” he says, but “another part is what I think of as maximising my luck surface area.”

This means creating as many opportunities as possible for surprisingly good things to happen:

  • Write publicly.
  • Reach out to researchers whose work you admire.
  • Say yes to unusual projects that seem a little scary.

Nanda’s own path illustrates this perfectly. He started a challenge to write one blog post per day for a month to overcome perfectionist paralysis. Those posts helped seed the field of mechanistic interpretability and, incidentally, led to meeting his partner of four years.

His YouTube channel features unedited three-hour videos of him reading through famous papers and sharing thoughts. One has 30,000 views. “People were into it,” he shrugs.

Most remarkably, he ended up running DeepMind’s mechanistic interpretability team. He’d joined expecting to be an individual contributor, but when the team lead stepped down, he stepped up despite having no management experience. “I did not know if I was going to be good at this. I think it’s gone reasonably well.”

His core lesson: “You can just do things.” This sounds trite but is a useful reminder all the same. Doing things is a skill that improves with practice. Most people overestimate the risks and underestimate their ability to recover from failures. And as Neel explains, junior researchers today have a superpower previous generations lacked: large language models that can dramatically accelerate learning and research.

In this extended conversation, Neel discusses all that and some other hot takes from his four years at Google DeepMind. (And be sure to check out part one of Rob and Neel’s conversation!)

This episode was recorded on July 21.

Video editing: Simon Monsour and Luke Monsour
Audio engineering: Ben Cordell, Milo McGuire, Simon Monsour, and Dominic Armstrong
Music: Ben Cordell
Camera operator: Jeremy Chevillotte
Coordination, transcriptions, and web: Katy Moore

Continue reading →

80,000 Hours review: 2023 to mid-2025

Blog post by · Published September 15th, 2025

Introduction

This review is structured in two parts: a summary of major organisational updates since our last external review (which covered up to the end of 2022), and a snapshot of where each of our programmes is at right now.

Given that we’re not currently fundraising, we initially considered not publishing an external review this year so that we could focus on other priorities. However, since we think these reviews might be informative for our audience, the EA/AIS community, and other stakeholders, we decided to publish a minimal version instead — we’ll stick to backward-looking updates rather than opinionated reflections or future plans, and largely draw on resources we’d already produced for other purposes.

The focus on brevity also means it’s more positively framed than it otherwise would be — both because challenges are harder to write about clearly, and because the programme updates are primarily focused on developments from this year (which have been going relatively well so far).

High-level org vision

80,000 Hours provides research and support to help talented people move into careers that tackle the world’s most pressing problems. We’re currently focusing our proactive effort on helping people work on safely navigating the transition to a world with powerful AGI, since we think this is the most pressing problem. We are broadly trying to:

  1. Be a great source of information to bring people up to speed on how and why to use their careers to make AGI go well
  2. Build automated systems for getting people into the roles that help make AGI go well

To do this,

Continue reading →

    Technical AI safety upskilling resources

    Blog post by · Last updated September 15th, 2025 · First published June 4th, 2025

    Sometimes, our advising team speaks to people who have enthusiasm for technical AI safety and a related skill set but need concrete ideas for how to enter the field. This list was developed in consultation with our advisors to find the resources they commonly share, including articles, courses, organisations, and fellowships.

    While we recommend applying to speak to an advisor for 1-1 tailored guidance, this page gives a practical, non-comprehensive snapshot of how you might move from ‘interested in technical AI safety’ to ‘starting to work on technical AI safety.’

    Overviews:

    Courses:

    Continue reading →

    #222 – Neel Nanda on the race to read AI minds (part 1)

    Podcast by · Published September 8th, 2025

    We don’t know how AIs think or why they do what they do. Or at least, we don’t know much. That fact is only becoming more troubling as AIs grow more capable and appear on track to wield enormous cultural influence, directly advise on major government decisions, and even operate military equipment autonomously. We simply can’t tell what models, if any, should be trusted with such authority.

    Neel Nanda of Google DeepMind is one of the founding figures of the field of machine learning trying to fix this situation — mechanistic interpretability (or “mech interp”). The project has generated enormous hype, exploding from a handful of researchers five years ago to hundreds today — all working to make sense of the jumble of tens of thousands of numbers that frontier AIs use to process information and decide what to say or do.

    Neel now has a warning for us: the most ambitious vision of mech interp he once dreamed of is probably dead. He doesn’t see a path to deeply and reliably understanding what AIs are thinking. The technical and practical barriers are simply too great to get us there in time, before competitive pressures push us to deploy human-level or superhuman AIs. Indeed, Neel argues no one approach will guarantee alignment, and our only choice is the “Swiss cheese” model of accident protection, layering multiple safeguards on top of one another.

    But while mech interp won’t be a silver bullet for AI safety, it has nevertheless had some major successes and will be one of the best tools in our arsenal.

    For instance: by inspecting the neural activations in the middle of an AI’s thoughts, we can pick up many of the concepts the model is thinking about — from the Golden Gate Bridge, to refusing to answer a question, to the option of deceiving the user. While we can’t know all the thoughts a model is having all the time, picking up 90% of the concepts it is using 90% of the time should help us muddle through — so long as mech interp is paired with other techniques to fill in the gaps.

    In today’s episode, Neel takes us on a tour of everything you’ll want to know about this race to understand what AIs are really thinking. He and host Rob Wiblin cover:

    • The best tools we’ve come up with so far, and where mech interp has failed
    • Why the best techniques have to be fast and cheap
    • The fundamental reasons we can’t reliably know what AIs are thinking, despite having perfect access to their internals
    • What we can and can’t learn by reading models’ ‘chains of thought’
    • Whether models will be able to trick us when they realise they’re being tested
    • The best protections to add on top of mech interp
    • Why he thinks the hottest technique in the field (SAEs) are overrated
    • His new research philosophy
    • How to break into mech interp and get a job — including applying to be a MATS scholar with Neel as your mentor (applications close September 12!)

    This episode was recorded on July 17 and 21, 2025.

    Video editing: Simon Monsour, Luke Monsour, Dominic Armstrong, and Milo McGuire
    Audio engineering: Ben Cordell, Milo McGuire, Simon Monsour, and Dominic Armstrong
    Music: Ben Cordell
    Camera operator: Jeremy Chevillotte
    Coordination, transcriptions, and web: Katy Moore

    Continue reading →