I Went On A Hunger Strike Outside Google To Stop The AI Race

I went on a hunger strike in front of Google Deep Mind, one of the biggest AI companies in the world. >> I was almost passing out. >> How the hell did I end up here? By now, you've probably seen the headlines about how the world's most cited AI researchers are worried that superhuman AI could kill us all. >> So, I actually think the risk is more than 50% [music] of the existential threat. >> But AI companies are still racing ahead. This year, Google execs said they need to move faster as a company. Even though the CEO of their AI division, the Mrs. Sabis, sign a statement in 2023 saying that mitigating the risk of extinction from AI should be a global priority. But how did I end up on British TV trying to talk to the Mr. Zabis, the CEO of a billion dollar corporation? Let me explain. My name is Michael Trazzy and I make documentaries about the AI race. In high school, I used to play the game of Go competitively and at age 18, I traveled to Japan to train [music] with professionals. Back then, Japanese professionals would make fun of AI. So, in 2015, when DeepMind released an AI that crushed a professional for the first time, I was shocked. It beat Fan Hui who was the top European Go champion for [music] three years running and I had the privilege of standing under him when I was 16 years old. So when Deepmind's AI beat my former teacher, it felt personal. >> Alpha Go win all the game. >> Then they hired him to help train their AI and several months later, Alpha Go beat Lee at all. Widely regarded as one of the strongest Go players in history and also happened to be my favorite Go player. It seemed like no one had seen it coming until I discovered a book Super Intelligence. The author Nick Bostonramm had predicted that AIS would one day surpass humans at virtually everything. Having personally witnessed that with the game of Go, where humans had been considered unbeatable. I believed it. But super intelligence made another huge claim that super intelligent AI could lead to human extinction. The argument was simple. If humans were to build machines that were better than us at everything, what guarantees do we have that they would still care about us? After all, when humans create a new house, do they really ask the ants for permission beforehand? So, in 2019, I went to Oxford to visit the man who wrote super intelligence and to work on AI safety. At the time, it seemed like Deepmind cared about AI safety. Deepmind co-founder Shane Le who had said in the past that they thought there was a 5 to 50% chance that AI could cause human extinction would come visit us in Oxford every other month to talk about AI safety. It seemed like they were taking the proper approach. But all that changed in 2022 when Chpt came out. Google realized they were losing to OpenAI and they declared a code red calling back Google co-founders Larry Page and Sergey Breen. Sergey left behind his yacht to take the lead on Gemini, Google's competitor to Chat GPT. The result was a corporate power struggle with bittersweet results. Deis Sabis got to remain CEO of Google DeepMind. But the mission had changed. Deis's DeepMind, which for years had been run like an academic institute, was now a corporate tool focused on shipping AI products like the Gemini chatbot. To me, it seemed like the miss Sabis was now compromising the founding principles of DeepMind [music] so that Google could chase money. But why did I go on a hunger strike outside their office? You may have seen the headlines about how 60 UK lawmakers accused Google of breaking its safety pledge by rushing ahead and releasing their chatbot Gemini 2.5 Pro. Google decided they couldn't wait a few weeks for proper safety testing. They chose winning the air race over safety. The UK lawmakers called it a dangerous precedent. The message was clear. Google shouldn't be blindly trusted with humanity's future. So on June 30th, 2025, a group called Pause AI protested outside Google Deep Mind, asking them to honor the safety commitments they made last year in Seoul. I was excited to see that people were finally taking this problem seriously. But on the day of the protest, we spent less than an hour in front of the demand office and then everyone went home. It left me feeling frustrated. A single hour wasn't enough. We needed something that would last weeks, months, or even years, bringing pressure on companies that would be impossible to ignore. So, I kept talking to people at PAI about protesting regularly in front of the demand office. But the motivation for a longer protest wasn't there. I was beginning to lose hope. But overnight, all that changed because of one person. On September 4th, a man named Guido Rexader posted on X that he was on day three of a hunger strike in front of AI company Anthropic, calling on employees to do everything in their power to stop the AI race. Of the big four AI companies, Entropic is widely considered to be the most safety conscious one. Their founder and CEO, Dario Amade, even testified before the US Senate, warning about new biological weapons that could be created with AI. >> We believe this represents a grave threat to US [music] national security. >> Even worse, when asked about AI causing human extinction, he said, >> "So, I think there's a 25% [music] chance that things go really, really badly." >> But Dario still describes himself as an optimist. a 25% chance of things going very badly for him means that there's a 75% chance that things could go really really well. Personally, I don't find that reassuring. Weirdly enough, Daario left OpenAI and fun anthropic in 2021 because he wanted to make a safer AI company, but he's now one of the main AI CEOs racing ahead, which according to his own words could lead to human extinction. That contradiction is why Guido was on a hunger strike in front of Enthropic. >> Yeah, it's um it's the fourth day that I'm out in front of Anthropic. I [music] just watched an interview with Dario. He reiterated this view that the race [music] to AGI super intelligence is impossible to stop. It's inevitable. [music] This is a similar view that Sam Alman has expressed. It's inevitable. There's [music] nothing we can do to stop it. And this is a lie. It's a very [music] selfserving lie. >> Guido's hunger strike gave me hope. So, a few hours after saying Guido's post on X, I decided it was time to finally execute my plan. I would show up in front of Google Deep Mind every day going on a hunger strike. Roughly what happened is I stopped eating on Friday. I knew hunger strikes could be deadly, but I believe it was worth the risk if we could pressure Google deep mind to stop racing to super intelligence. So my ask for the hunger strike would be for Demis to publicly state that it would stop releasing more powerful models if all the other major labs agreed to stop too. To get to that goal, I would need to keep the protest going while gathering help from the press and even Deepmind's own employees to put Deis in a situation where he would be unable to say no. On day two, [music] I run into a problem with King's Cross Security, responsible for the private land that Deep Mind's office is on. They told me that my Sharkboard was a safety concern. It sounded like a phony reason to me, but I was standing on private property and didn't want to be kicked out. So, I gave in and took down my sign like they told me to. I left Dubine's office feeling discouraged. But help was on the way. As I was finishing my second day, someone back in Amsterdam was preparing himself [music] to join me the next day. This guy, Dennis, >> my name is Dennis Sherid, >> who I had never met before, was willing to fly all the way from Amsterdam to join me in my hunger strike in London. I learned later that Dennis had been pursuing a master's degree in AI at Amsterdam, but dropped out before finishing his thesis. He had realized the main problem wasn't technical. It was that companies were racing to build super intelligence as fast as possible, and no amount of safety research was going to stop them. So in 2025, he co-ounded together Against AI, taking a more direct approach to warning the world. So the next morning, we had two different signs. I was on day three of my hunger strike, and he was on day one without eating. But then King's Cross security came to see us again. Google wasn't happy about us having multiple signs saying hunger strike in big letters. But we decided [music] to stand our ground. If we didn't have any signs, what was the point of being there? Without the signs, no one would even understand what we're protesting or why. The security guy called his boss and we waited, unable to hear what was being said on the other end of the line, unsure of whether this would be the end of our hunger strike. Finally, the security officer told us that we could stay, we could keep our sign. We were relieved that their protest would continue. But for the first 3 days, we had still not talked to a single Deep Mine employee. If nobody at all was willing to talk to us, what hope did we have of reaching CEO Deisabis? But the next day, things finally started to turn around. We started to really see more attention from the press. There were two pieces on business insider and journalists came first thing in the morning to talk to us. We were really encouraged with the press covering a protest. We knew the deis couldn't keep ignoring it. And then on day five, something unexpected happened. Our day started simple enough. We had a remote interview with John Sherman for his for humanity podcast, which featured an interview with Guido, [music] calling anthropic CEO Dario Amod to stop participating in the air race immediately. If he's made a conscious decision to do something that could end up killing my children, I've got a right and I've got a responsibility to talk to the person in charge. >> Talking to [music] the person in charge was also our goal. But how could our voices reach the CEO of one of the top three AI companies in the world? It seemed hopeless, like we were screaming into the void. But that evening, one very familiar face was leaving the deep mind office. This was David Silver, principal research scientist at DeepMind. He was featured in an award-winning documentary about DeepMind as one of the key researchers behind Alpha Zero, Alpha Star, and Alph Go. Yes, that Alph Go, the one that had been my French Go teacher and my personal hero. [music] The same AI that had driven me to start studying AI safety. We talked for a long time about whether Deis was even able to talk about pausing AI and what good outcomes would look like for him. I asked David if he would be willing to deliver a letter to Demis containing our specific ask. He took our letter and said he would try. So on that night I went to sleep hoping that our letter would reach him. And the next day we received confirmation that a letter had been forwarded directly to Demis. Deis had read our letter. He knew what he wanted from a public statement. The question was would he actually agree? But day six was also when the strike [music] started to really take a toll on me. >> I think maybe this the first day where I'm like really feeling the hunger strike. It's been affecting my sleep. It's been affecting how much I exercise. >> And by the next night, things would get dangerous. [music] >> Today is Friday, September 12th at midnight, 49 minutes. Um, over the past uh 48 hours, um, I've [music] had a couple moments where I felt like I was almost passing out. A couple hours ago, I received a message from two doctors that [music] told me that I had um very low glucose levels, even low for people [music] doing uh, you know, a long fast within hours. That could lead to seizures, could send me into a coma or even kill me. It sounds dramatic, but that's what severe hypoglycemia can do to the human body. [music] But stopping Mand Dennis would have the entire burden of keeping the hunger strike going alone. After getting a letter to Dennis, after all the media coverage, quitting now felt like throwing it all away. But the doctors were clear continuing could kill me. So I spent hours that night weighing my options, trying to estimate the risk of me dying overnight. Given my glucose levels [music] were dangerously low and multiple doctors were telling me to stop, I made the hardest decision of the strike and probably of my entire life. I had to stop. My goal before going to sleep is to uh increase my glucose levels by by a little bit which means I'm breaking my fast. Um for a lot of people eating is is you know part of their life but breakfast at noon the dinner is something we do all the time. Uh but for me this feels like something I haven't done in a very long time [gasps and sighs] and I was forgot what it is to eat. So in that notes and this marks the end of the fast. Today is um day six of the hunger strike for Daniel. How I feeling like optimistic, pessimistic. Yesterday it was raining most of the day, so I was just hiding from the rain, basically sitting under an umbrella. >> How was it? >> Well, I have uh I have a lot of layers of clothes, so it wasn't that bad, but it was it felt pretty depressing. >> That day was Dennis's first day protesting alone. And I knew from personal experience how demoralizing that was. And things were about to get even worse for Dennis. >> Today is day seven of my hunger strike. It's been getting pretty difficult recently. Yesterday I went to bed and then I was just thinking about burgers for like three hours straight. It felt so tiring to come here. Um I felt so fatigued and I just had to take also a couple of breaks uh from just walking. >> Dennis was alone in London, but at least he had a friend also on a hunger strike outside the offices of Air Company Anthropic in San Francisco. Hey man, how's it going? >> All right, here I >> So Dennis was still getting some support and in fact Max Techmark the [music] MIT professor behind the original pause statement that many people signed in 2023 also reached out with words of encouragement. >> One very powerful signal that you're sending to people know about what you're doing is that this is serious. If it wasn't something serious, you wouldn't be doing this. a lot of scaredy-cats working in these companies who are just so afraid of that their boss is going to look at them funny and just just get the even the most minor social pressure and that's going to make them fold and not say anything. So, so hopefully it's very inspiring for them to see that here's someone who's willing to risk a lot more than a little bit of social stigma for follow their their moral compass. >> But then support started coming from an unexpected direction. A couple people from Deep Mind walked up [music] to me and then they gave me a thank you card. Uh, and it says, uh, thank you very much for your action. We really appreciate the work you do and want to let you know that it got a lot of people talking in deep mind. We hope you get an answer from Demis or Deep Mind. Do take good care of yourselves. And then they signed it. So, uh, this was pretty cool >> cuz it shows that there's like debates and conversations happening. So it's not like just newspaper, it's also people like inside of Deep Mind having conversations [music] and maybe shaking things up internally. On top of that, one of my connections at Deep Mind told one of my friends that [music] the hunger strike had been way more impactful than he had expected, even if we couldn't see the impact from outside the office. Our hunger strike was making a difference. The Deep Mine employees were on our side, even if they couldn't say so publicly. But the next day, the Deep Mind finally broke their silence on the hunger strike. They told The Verge, "Safety, security, and responsible governance are and have always been top priorities." They say in corporate interviews that they care about safety, security, and responsible governance. They claim that safety is their priority, but they keep racing ahead. So, we wrote back, "If they actually cared about safety, they would have spent a couple more weeks doing safety testing before releasing Gemini 2.5 Pro. they wouldn't have broken their pledge and created what 60 UK lawmakers called a dangerous precedent. Google's statement was a step forward, but we needed them to actually agree to stop racing if everyone else also stopped. And the next day, our movement kept gaining momentum. We had a piece published about us in Lemon, the French equivalent of the New York Times, and we also appeared on TV in the UK. I don't think it's that of an extreme of a response because the CEO of Google DeepMind has said that there's a non-neglable chance that uh the technology they're developing will literally lead to human extinction. >> The hunger strike had been a success beyond our wildest dreams appearing on TV, getting major press, getting our message directly to the CEO, and even forcing Google Deep Mind to make a public statement. Think about it. were just two guys with a sign and a billiondoll company had to respond to us. That meant the public pressure was working. But as the hunger strike continued, the question now was how far could Dennis and Guido go before their bodies hit their breaking point. So today we're on day 15 of Dennis without eating. How are you feeling today, Dennis? >> I've lost [music] quite a lot of weight. Yeah, it looks like he just came back from the war or something like >> Well, I was thinking of stopping yesterday, but the Telegraph and Sky News published a piece about us. >> Like you feel like you've exhausted all your strength. >> Uh I I don't really know. I'll I'll see. >> The next day, we reached a major breakthrough. Our story in the Telegraph was published in their print edition as a major news story. We were really having an impact. But after more than two weeks without food, Dennis seemed to be at the end of his rope. >> Okay. So, it might be the last one, right? >> You were 50. Now you're like smiling. You're like incipation. >> Yeah. I mean, tonight I stayed up till like 5:00 a.m. just videos. Yeah. About like cooking recipes. >> So, you had terrible sleep. Why did you sleep 6 hours? Now more like four. >> Oh >> Yeah. I I really really really want to eat again. Like at the moment I'm not thinking about it that much, but then I I know that like when I get home tonight I will be just thinking about food the whole time. Okay. So after 16 days of not eating, I'm going to eat for the first time. Let's see how it tastes. Yeah, this really good. I'm not sure I would like this if uh I wasn't so hungry, [music] but uh yeah, this I like this. >> Dennis's hunger strike was coming to an end, but Guido, the man what inspired me to start, was still going on day 23 of his own hunger strike in front of a company Anthropic in San Francisco. How are you feeling, Guido? After 30, actually 31 days of not eating. >> 32. >> About 2 weeks later, Guido Rexetter finally announced the end of his hunger strike against AI company Anthropic. But this is only a very small step in humanity's journey towards building safe and beneficial AI. It's not only me, Dennis, and Guido saying this. The top three the most cited AI scientists have expressed that they are worried that AI might pose an existential threat. And most recently, over 100,000 people, including Apple co-founder Steve Wnjak and even Prince Harry, sign a statement calling for a ban on building super intelligence until it can be done safely. [music] Damis and Daario know there's a risk that AI could lead to human extinction. Yet, they keep racing to build this potentially worldending technology. The problem is not that top AI CEOs do not understand the risk. It's that they're trapped in a deadly air race where there are no winners. But we won't end the race by convincing one or two CEOs. We need governments around [music] the world to realize the danger and enforce agreements that will affect all AI companies. And that's the most important thing we accomplished with the hunger strike. The Mr. Sabis may have not listened to me, but the media and the public did. An Axio's poll from earlier this year showed that 77% of Americans want AI companies to build slowly and safely rather than recklessly racing ahead and 87% of Britain want [music] a safety first approach to new AI models. We already know the dangers. We just need to stand up and demand that our governments take action to stop the AI race. All of this happened because three people decided to protest in front of AI companies entropic and deep mind. So what could 100,000 people do? Scan this code today to sign the statement calling for a ban on super intelligence. And if 100,000 people pledge to march in Washington DC calling for an international treaty, we will all be marching together.