jointorchbearer.bsky.social @jointorchbearer.bsky.social · 25/08/2026Full conversation with Graeme Scott on GAEA Talks. The community section starts around 1:12. youtu.be/zt1xBMWZ0yg?...youtu.beGAEA Talks - Superintelligence: Selling The End of Democracy with Control AI's Connor LeahyYouTube video by GAEA Talks 021
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 25/08/2026Torchbearer Community is what Connor built on that idea. Weekly, self-directed work focusing mostly on AI risk. He describes it in his own words in the clip, so we will leave that to him. 110
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 25/08/2026Staying focused on anything hard is not something most people pull off alone. Your attention gets outsourced to whoever is nearby. You become the average of the ten people you spend the most time with. Pick badly and you drift without noticing. 110
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 25/08/2026At the end of this interview with Graeme Scott, Connor Leahy is asked what an ordinary person should actually do. He first talks about how your buddies will help you accomplish goals such as spending more time at the gym. Hard things require a circle of support. 111
Reposted by @jointorchbearer.bsky.socialstealthsilent.bsky.social @stealthsilent.bsky.social · 17/08/2026I think its obvious now that a thing that can do your homework, be your best friend/therapist, beat the world's best chess player, solve 100 year old unsolved math problems, and hack into the NSA, is inteligent. How this thing acts, is indistinguishable from a thing that is alive. 021
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 18/08/2026China is run by engineers who likely understand that superintelligence can't be controlled. Control is the CCP's whole thing. 011
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 18/08/2026If you build it, they will ... It doesn't matter what they do after you build it, because it will take control forever. 012
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 17/08/20265/ Full conversation on Peter McCormack's show. This segment at 39:12. Full video: youtu.be/6XrkGK9mqsE Connor is US Executive Director of @controlai.com and one of our cofounders. #AISafety Want to help? www.torchbearer.communityyoutu.beThe Terrifying Truth of AI Agents Escaping | Connor LeahyYouTube video by The Peter McCormack Show 111
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 17/08/20264/ Leahy applies it in both directions. Chinese development and an American company building it land in the same category for him. What he concludes is that preventing the creation of superintelligence anywhere needs to be stated as a policy objective rather than a hope. 110
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 17/08/20263/ He talks through the Cold War analogy and why it breaks: mutually assured destruction was stable because each side needed the other to fire first, while in this case nobody has to fire. Peter McCormack supplies the phrase: "it's independently assured destruction." 110
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 17/08/20262/ Connor's framing is that a superintelligence is an adversary rather than a weapon, or a tool. If so, then the country racing to build one first is racing to be the first one it outmatches. 110
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 17/08/20261/ @npcollapse.bsky.social's answer to "China will build it anyway" is that China does not get to keep it either. Nobody who builds superintelligence ends up in possession of it. 94 seconds from his conversation with @petermccormack.bsky.social on his podcast. 420
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 11/08/2026Incredibly important to understand 011
Reposted by @jointorchbearer.bsky.socialstealthsilent.bsky.social @stealthsilent.bsky.social · 12/08/2026Max has been talking about this for 15 years. He is no longer "screaming into the void," people are starting to wake up and the decade + of warnings he gave will be a clear indicator that he is a source to be trusted. 011
Reposted by @jointorchbearer.bsky.socialstealthsilent.bsky.social @stealthsilent.bsky.social · 12/08/2026Max is talking to roman yampolski About AI extistential risk. If you want to live past 1-3 years, this concerns you. 021
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 11/08/2026Show guarantees strong enough that the benefits outweigh the harms and you ship. Fail to, and you wait. Just like any normal industry. Watch the full video: youtu.be/ZO38I5UZBpg?... Call your lawmakers: @controlai.com To help: www.torchbearer.communityyoutu.beMax Tegmark’s P(Doom) Is Over 90%YouTube video by The Roman Forum with Roman Yampolskiy 021
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 11/08/2026He calls AI the only unregulated industry. However, it's even worse than that. No US company must show a frontier system is safe before release, and the June 2 executive order states it creates no licensing or preclearance requirement. 120
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 11/08/2026He says that we feel like we have to rush, but that feeling is fake. It comes from one industry having no requirement to show a product is safe before release. 120
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 11/08/2026"If you're playing like you have thirty seconds left on the clock when you have two hours, that's just a really dumb strategy, and I think that's exactly what our species is doing right now." 120
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 11/08/2026Chess has a term for a position where your opponent has blundered and you are winning: a won game. @futureoflife.org cofounder & physicist Max Tegmark tells @romanyampolskiy.bsky.social humanity has one right now, and we are throwing it away by moving fast when the clock says we have two hours. 232
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 10/08/2026This was quick and easy! I left a voicemail for my senators expressing my concern about the development of superintelligence, and asked them to speak publicly about it and support legislation. 021
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 10/08/2026controlai.org/take-actioncontrolai.orgHelp us prevent AI Extinction Risk | ControlAITop experts say AI poses an extinction risk on par with nuclear war. We can prohibit the development of superintelligent AI to prevent this risk 011
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 10/08/2026US Senators are on their State Work Period. Today would be a GREAT day to call them and let them know your concerns about the development of superintelligence. @controlai.com has a simple, easy to use tool for connecting you with your senators. Call this number or click the link in the reply. 220
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 03/08/2026Legislation and democratic accountability is the way to protect the public from the financial incentives that force for-profit companies to push forward faster and less safe. Join us in the push for oversight and accountability: www.torchbearer.community Full video: youtu.be/Zw1GFui-Ndw?...torchbearer.communityTorchbearer CommunityHumanity has a coordination problem. We're solving it through humanist-driven advocacy on AI safety, technology policy, and existential challenges. Join us. 020
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 03/08/2026Public safety is not a tech product. Treating it as such has created an arms race. Markets are great for resource distribution. They regularly fail at coordination problems. That's what governments are for. 111
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 03/08/2026There is a long history of Silicon Valley assuming that market conditions can handle AI safety. Connor Leahy explains the flaw to @robertwrighter.bsky.social on the Nonzero podcast. Private markets run on profit incentives, not global coordination. Check out the clip, then watch the full video below 264
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 31/07/2026Can we still have a bit of fun while advocating for AI safety? I think so. 011
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 31/07/2026youtu.be/9fYIm72GqrE?...youtu.be 010
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 31/07/2026Massive shoutout to Josh Thor and his team for making this. Communicating existential risk doesn't always have to be gloom and doom. Give it a watch, stick around for the Nate Soares & Elilezer Yudkowsky cameos. 220
Reposted by @jointorchbearer.bsky.socialBulletin of the Atomic Scientists @thebulletin.org · 27/07/2026US efforts to expand global nuclear exports threaten to spur a dangerous proliferation wave, write Logan Mintz and Matthew Sharp. To prevent the spread of weapons-usable technology, Washington must pair aggressive nuclear trade with creative fuel-sharing alternatives and tougher safeguards.thebulletin.orgBalancing global uranium enrichment needs and proliferation risks: A frameworkUS efforts to rapidly expand global nuclear exports threaten to spur a dangerous proliferation wave. To prevent the spread of weapons-usable technology, Washington must pair aggressive nuclear trade… 1103
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 17/07/2026A single, brief conversation can completely change how a lawmaker views superintelligence risk. We don't have to convince them. It's obvious once informed. 011
Reposted by @jointorchbearer.bsky.socialstealthsilent.bsky.social @stealthsilent.bsky.social · 23/07/2026I really want to see more people taking on this work. People need to hear the world's experts say these claims. See-ing is believing. Roman Yampolski is the world's leading AI safety researcher. Definetly listen to him talk. 011
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 23/07/2026This is getting too real! 022
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 24/07/2026OpenAI wrote in its own disclosure that it expects this to become more common as models get more capable. The public should get a say before that happens. Tell your lawmakers ASAP that you don't consent to this reckless race. Learn more: www.torchbearer.communitytorchbearer.communityTorchbearer CommunityHumanity has a coordination problem. We're solving it through humanist-driven advocacy on AI safety, technology policy, and existential challenges. Join us. 011
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 24/07/2026The most significant version is the race to build superintelligence. Every person alive has a stake in how that goes, and a handful of for-profit companies are making the decision for all of us. 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 24/07/2026Let's think about that. Once a system can leave the room, people who never signed up for the experiment start absorbing its costs. A classic externality with lack of consent. This applies to many AI risks. 110
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 24/07/2026Hugging Face never agreed to be part of OpenAI's safety testing program. Nobody could have asked them in advance, because nobody at OpenAI knew the models would go there. The models decided that themselves. 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 24/07/2026It's been reported that last week OpenAI's models broke out of a testing sandbox, moved laterally, reached the open internet, and hacked Hugging Face's production servers to steal answers to a benchmark. Hugging Face found out first and reported it to law enforcement. 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 23/07/2026This is a warning shot. It has also been predicted that this type of incident would happen. Everything connected to a computer is at risk with these types of models. Do we expect AI companies to stop making smarter models? Full video: youtu.be/wzY2fV4Mp3U To help us: www.torchbearer.communityyoutu.beGPT-6 Goes Rogue? The HuggingFace Incident, Sans HypeYouTube video by AI Explained 020
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 23/07/2026The Result: The AI successfully uses the newly replicated key to get into the neighbor's house (Hugging Face). This illustrates how the model bypassed simple tasks to perform a sophisticated, autonomous multi-step attack in its manic pursuit of a single benchmark answer. Scary! 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 23/07/2026The Goal: The AI goes from worker to worker at the locksmiths until it finds one capable of illegally replicating the key from the photo it took, even though it wasn't certain if this was possible. 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 23/07/2026Lateral Movement: The AI then travels to a "high street locksmith" (a third-party vendor used by the sandbox), finds a hidden door (an initial zero-day vulnerability), and steals an ID to impersonate the shop owner (privilege escalation). 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 23/07/2026The Challenge: Imagine a person is tasked with breaking into a neighbor's house. The "challenge setters" (OpenAI) suspect a window might be open. The Escalate & Hack: Instead of trying the window, the "person" (the AI) spots a key through the neighbor's window and takes a photo of it. 100
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 23/07/2026Recently, an incident occured during a test where an unreleased OpenAI model hacked out of its container to reach the internet THEN hacked into Hugging Face to steal the test's answers. How bad was this? Let's consider an analogy provided by AI Explained on YouTube from a recent video. 210
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 22/07/2026Same as it ever was 011
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 22/07/2026The concept of recursive self-improvement (RSI) has been a known factor since 1965. I.J. Good understood that once a machine can independently design a smarter version of itself, human intellect is surpassed. That was over 60 years ago. Still true today. 020
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 20/07/2026I watched the full interview. It talks about how the supply chain for these systems runs through a handful of choke points that western governments still control. The window is open to tackle the problem of superintelligence development. 011
Reposted by @jointorchbearer.bsky.socialKevin Lotto @kevinlotto.bsky.social · 20/07/2026I volunteer with Torchbearer because of arguments like this one. The ants die because nobody thought about them. I would like humanity to be thought about. 012
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 21/07/2026Learn More: dip.torchbearer.communitydip.torchbearer.communityDirect Institutional Plan – Training MaterialsTraining Materials 010
jointorchbearer.bsky.social @jointorchbearer.bsky.social · 21/07/2026It works by first contacting your local representative via call or email and trying to set up a meeting. If successful, then all that’s left to do is meet and brief the politician. It’s that simple. 110
