The AI Wealth Problem

Anthropic is proposing to give away hundreds of billions of dollars in public benefit funding. So everyone's pretty excited about this, right? In fact, it's bringing up a lot of interesting, important, and sometimes difficult questions about not only the opportunities but also the challenges of AI wealth, as well as the relationship between highly concentrated AI wealth and power and the public sector and democracy at large. These are poised to be some of the biggest questions we face as a society in the coming years. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Robots and Pencils, ZData, and Blitzy. To get an ad-free version of the show, go to patreon.com/AIDailyBrief, or you can subscribe on Apple Podcasts, And if you want to learn more about sponsoring the show, send us a note at sponsors@aidailybrief.ai. We start off today with a story that many are paying attention to because of what it potentially reveals about the world we now live in. OpenCode suffered a major cyberattack that could undermine support for open-source alternatives to the big labs. On Friday night, users saw linked API accounts running up massive bills in the hundreds or even thousands of dollars in some cases. By Saturday afternoon, OpenCode acknowledged the attack, writing, we had a security breach last night related to a large-scale distillation attack that impacted 1.4% of OpenCode Console users. They said that the attack was still ongoing at that point, and they were working with OpenAI, Anthropic, and Google on remediations. If you're not familiar with their service, OpenCode offers an open-source coding harness akin to Codex or Claude Code. Users typically bring their own models, storing API keys with OpenCode. This was the attack vector, with the full database of user API keys seemingly exposed. Even OpenCode founder Dax Rad was vulnerable. On Saturday afternoon, he posted, my personal accounts are being attacked now too. Whole new world with AI-powered hacking. OpenCode warned users that they need to rotate their API keys, not just revoke access. Anthropic said that they were mitigating the damage on their end and were revoking any compromised keys that they could detect. And as the weekend continued, we got some additional details about what the attack involved. On Sunday morning, Dax posted, they've been nonstop trying to attack our systems for the past 48 hours. A side effect is everything is getting hardened, but damn, we basically need agents going 24/7 because that's what they have. This sounds similar to the Hugging Face incident, where a swarm of agents pounded the servers for several days. The Hugging Face agents rapidly spun up infrastructure to avoid shutdown, moving much faster than human defenders could. Like Hugging Face, Dax said he ran into the issue of Frontier Model Labs blocking cyber tasks, leaving him reliant on open-source models. He wrote, that was a bit frustrating mid-incident when trying to investigate. They did quite a bit and then they started refusing. We got access to Daybreak Blue now, so it's better. Daybreak Blue is OpenAI's vetted cyber defender program, and Daxetti is now running continuous defensive agents using the service. Overnight on Sunday, Thibault from OpenAI confirmed that damage mitigation had concluded on their end, with API keys reset and refunds issued to impacted customers. The story is still evolving, and presumably we'll have a lot more information later in the week. But at first glance, there's reason to think that this incident could have a very significant impact. It appears to be the first instance of an agentic swarm carrying out a large-scale malicious attack. Now, OpenCode said that the hack was related to a distillation attack, which is causing a little confusion on X. It seems that user API credentials for OpenAI, Anthropic, and Google were used to generate a ton of tokens very quickly for later use as synthetic training data. That doesn't necessarily imply a Chinese lab was behind the attack, with API access potentially being sold through black market resellers. However, it does imply the hackers were able to directly monetize the attack by stealing valuable digital services rather than just digital information. The attack could also have big implications for how AI companies deal with security. Services that store API keys are now looking a lot more like financial services and will need to beef up their security, likely with always-on defensive agents. Former AI czar David Sacks used this moment to advocate for widely distributed cyber-capable models. He wrote, this illustrates the point I've been making about the limits of centralized approaches to SI safety in cyber. And yes, I apologize, for some time we're going to continue to have to deal with White House allies using the term SI while everyone else continues to use AI. But in either case, Sacks continues, the only durable response is decentralized. Make powerful models available so defenders can build and run their own cyber defenses. The only solution to SI-powered cyberattacks is SI-powered cyber defense. Dean Ball of OpenAI called the lack of access to Daybreak a, quote, fairly modest hindrance that was quickly resolved by being granted access to a trusted access program, so seems reasonably fine versus the cost of enabling sustained autonomous cyberattacks, question mark? And this is a pretty good proxy of the debate right now. Is the right approach specialized access granted by the labs who can whitelist someone as part of a program? Or is it just having the models more generally available? Now, another big potential implication is the swinging pendulum of open versus closed models. OpenCode is the leading open coding harness and could be suitable for enterprises looking to go multimodal. One user on the OpenCode Reddit said that they were pitching the service to their company, but the compliance department will likely now shut it down. Like I said, there will presumably be lots more to come. For now, we are officially in it. Welcome to the new era of cyber. Meanwhile, over in markets, Wall Street is on edge as analysts call an AI bubble burst. The past few weeks have been full of extremely bad news for the Neo Clouds. First, the failed Firmus IPO leading into a massive drop in CoreWeave bonds. Charles Gasparino of Fox Business has seen enough, writing in the New York Post, the AI bubble is about to burst and a big market correction is coming. The article was a list of comparisons to the end of the dot-com bubble, with Gasparino concluding, common sense and history teach us it won't last. The correction is coming to AI and all the companies tapping into its technology through big spending, much of it using debt. The main ingredient of any crash. If history is any guide, it's time to take your profits and not get greedy, or pray that whichever AI you're betting on will be a survivor of the crash that's coming. Now, I will leave it to you to decide what it means that that op-ed is being presented in the New York Post, but the signs of pain are more widespread than just the Neo clouds, who have taken an absolute beating over the past couple of weeks. Semiconductors are in the middle of their longest correction since 2006, with the Philadelphia Semiconductor Index down 8 days in a row for a total loss of 12%. This is all weighing on the major Nasdaq index, which lost more than 5% last week. Of course, there are still pockets of optimism in the middle of the drawdown. Nvidia CEO Jensen Huang told Bloomberg on Friday, all of this is going to pass, and the reason for that is very simple. It's going to be transformative. It's going to impact in a very positive way every single industry and every single country. There's no question it will continue, and we're at the beginning of that. Elon Musk used an AI-generated video to highlight that compute supply is still insanely tight. He claimed that SpaceX can't find spare capacity at AWS, Google, or anywhere else at the moment. ARK CEO Cathie Wood, who knows a thing or two about investing in tech hype cycles, isn't concerned with what she's seeing so far. Wood said, the fear out there actually is good. I'm talking about fear that AI, we're in a hype cycle. We do not believe we are, but the fact that people are afraid means we are not in a bubble. She added, while some people might say we're in the first inning, we would say we're still in the parking lot before we enter the baseball game. We're very optimistic about semiconductors generally, about cloud computing, neo-cloud computing, because we've barely begun in AI. Whatever innings we're in, inevitably comparisons to the dot-com are gonna be left wanting. Opinions differ on the true state of demand, but we're certainly not hearing about massive overbuilds and canceled spending. Mostly we're just seeing investors demand a little more to finance risky data center projects through the bond market. That's crimping valuations for neo-clouds, But that's very different to the dynamics that burst the dot-com bubble. Some more evidence of that comes from spot GPU rental prices, which are up 17 days in a row now. The question to ponder is what it means if NeoCloud debt funding dries up. And at this point, for those of you who want to go deeper in markets analysis, what I would recommend is to shy away from anyone who tries to use big overarching terms or weighty, if inaccurate, historical comparisons, and instead look for the people who are trying to get into the actual specifics. Meanwhile, Microsoft has pulled the plug on a multi-billion-dollar UK data center deal over energy delays. According to The Telegraph, Microsoft has canceled their contract to build a large-scale data center in Essex. The site has faced multiple delays, with local utilities most recently saying they don't expect to connect the facility to the grid until the 2030s. It was originally intended to come online next year. This is a huge blow to Nscale, who positioned themselves as the UK's sovereign data center provider. Microsoft represents 40% of their $103 billion in contracted revenue, although this site is believed to be a small portion of that amount. Still, Nscale has been attempting to go public in an IPO deal that would make them one of the most valuable AI companies in the UK. They had been seeking a valuation of $30 billion, but according to Bloomberg, a group of investors have been pushing management to settle for a valuation that's in the $16 to $19 billion range. That would represent not a huge jump over their last funding round, which reportedly valued the company at $15 billion in September. The contract being canceled is also a pretty big blow to market confidence. This is one of the first high-profile projects to get mothballed, and another high-profile NeoCloud generating some terrible headlines. To take the optimistic read, this cancellation isn't, again, about a lack of demand or other structural issues with the AI industry. Per the reporting, it seems to be a narrow case of a data center without a clear path to electricity supply. That could say something about the scarcity of capital— essentially that Microsoft can't afford to have their budget allocated to a facility that won't come online for several years. More broadly, it could indicate that we're past the frothy part of the data center buildout when hyperscalers were signing contracts with brand new companies with zero track record. For the sake of the full breadth of takes, we have on the one end of the spectrum from Ben Bridgeman, Microsoft still has substantial contracts with Nscale. The recent issue in the UK is about power availability, not a lack of demand for Nscale's capacity. Are we really surprised that building enormous data centers in the Arctic Circle presents logistical challenges? And then on the other end of the spectrum from Nobody Special on X, Nscale IPO not looking good here. How long until they start using the word contagion to describe what's happening to NeoClouds? Lastly today, one from the Anthropic IPO files. New York City Comptroller Mark Levine has urged Anthropic to pause their IPO for safety reasons. In a letter sent to Anthropic last week, Levine wrote, Anthropic and 5 other companies recently signed a quote-unquote morally binding accord with the White House, which is insufficient to meet the moment. Before going public, the company should prioritize addressing the myriad safety concerns surrounding AI threats and establish a governance regime that matches the seriousness of the concerns being raised by you, as well as industry researchers, policymakers, and the public. The demands of preparing for a public offering diverts resources and attention from this very necessary work. Now, if you're wondering why on earth should we care if the Office of the Comptroller of New York City is upset with Anthropic, that is a very reasonable question. The Office of the Comptroller is in charge of a range of different financial auditing and oversight functions for city government, but in this instance, the office wrote in its capacity as the investment advisor and custodian of the $325 billion New York City pension fund. In other words, this could be a representative opinion from a deep-pocketed investor and might reflect the type of thing that others are thinking about as well. Levine noted that Anthropic's IPO filings contained 80 pages of risk disclosures, more than twice the length of the section describing the business. The letter stated, these disclosures are not routine and indicate risks that are without precedent, let alone at a public offering of this scale. The comptroller is calling for a similar level of detail on Anthropic's plans for auditing and controls to deal with this risk. The letter also called attention to reported attempts to grant super-voting shares to the founding group, giving them contracted control of the company. Levine argued, a gap between control and accountability is especially concerning for investors given the unprecedented risks the company has already acknowledged. The letter was sent before Anthropic's Thursday blog, which discussed the appointment of Meador as the first external auditor with employee-like access to the company. However, the comptroller seemed to be asking for a far more detailed plan and a specific timeline. Given the gravity of Anthropic's own concerns, Levine called on Anthropic to pause their IPO so the company can fully address safety issues. Levine appeared on Bloomberg TV to discuss the Anthropic IPO, noting the company had opened discussions with his office following his letter. He wasn't able to characterize the discussion, but said it does appear they are under pressure from multiple sides right now. If they go public and then there is a catastrophic safety incident far beyond what we've seen right now, that would be terrible for their stock. Much better to take a little time right now and get this done. Levine is assuming the indexes will rush Anthropic in, as the Nasdaq 100 did with SpaceX earlier in the year. Some of the city's pension funds are invested in the major indices, so they would have forced exposure over a relatively short time frame. Levine noted there's risk for the markets either way, commenting, some people are making the opposite point, that if they delay their IPO, then that in itself could be a problem for the market. However, he continued, I worry about going public and then having an incident. It's beyond the scope of the headlines to get into, but why I think this is an interesting moment is that it dramatizes a lot of the issues that are going to come up. As I've said before, I have a very hard time thinking that people are going to be comfortable with Anthropic's plan for super voting shares. And I think you're going to see a lot more resistance to that. But more broadly than that, there's an interesting question of the risk of an incident happening after Anthropic or OpenAI is public and that dragging the market down versus the increased visibility and transparency we get with a public company. I think there's a fairly strong argument that you in fact want these companies public as soon as possible if we want more of a hand and an influence in determining their direction. In any case, that is going to do it for today's headlines. Next up, the main episode. A new study from KPMG and the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes. Researchers studied more than 500 early-career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating, and refining its outputs. These top performers, called AI amplifiers, weren't defined by what they knew alone, but by how they worked with AI. Learn more about what separates AI Amplifiers from everyone else at kpmg.com/us/aiamplifiers. The best teams don't have a single star carrying everyone else. They know their own strengths and each other's weaknesses and play to both. That's the team Robots and Pencils has built on purpose. Nobody there is grinding through busywork to pad a headcount number. People come for the hard problems and they stay because everyone around them is leveling up at the same time. In a market full of companies that are just trying to hire fast, that's worth a look. Check out robotsandpencils.com/careers. This episode is brought to you by CData. Companies rolling out AI keep hitting the same wall. Token spend is climbing, security is holding agents back, and no one knows how AI is truly being used. The root cause? AI is cut off from the systems and context your business runs on. That's why CData built Connect AI Gateway, one control point between your AI and your systems. Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more, each agent gets exactly the data and tools it needs. It governs down to the record and lets you finally see your org's complete AI usage in one place. Accurate answers, a fraction of the token cost, and a gateway that learns your business with every interaction. See how at cdata.ai. Every AI coding tool on the market does the same thing first: it starts writing code. Blitzy does the opposite. Before writing a single line, Blitzy spends days reverse engineering your entire codebase. Thousands of agents ingest millions of lines, mapping every dependency, every undocumented constraint, every architectural decision made over the last decade. The result is a dynamic knowledge graph that understands your software the way a principal engineer would after 30 years in the building. Other tools guess at context with grep searches and markdown files. Blitzy never guesses. It builds true understanding first, then delivers over 80% of entire software epics autonomously. Validated, end-to-end tested, production-grade pull requests. That's why Fortune 500 engineering teams trust Blitzy with the codebases that matter most. See for yourself at blitzy.com. That's B-L-I-T-Z-Y dot com. Welcome back to the AI Daily Brief. Today we are talking about how AI is going to change the world, but in this case, it's a very different aspect of that conversation. Anthropic has just announced what amounts to a significant philanthropic arm. Over the weekend, they wrote, we're committing a portion of Anthropic shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit. It will start off with 5% of our shares. For every 50% increase in the company's value over a set baseline, we can contribute an extra 2%, subject to our board's discretion. In total, this means that roughly 20% of Anthropic shares are committed to the public good. Our co-founders have already pledged to donate 80% of their equity, and around half of our employees are donating a portion of their equity as part of a donation matching program we've run since our founding. In the companion blog post, they go a little bit deeper into some of their theory of change and how it relates to the beneficial mission of the company as a whole. They write, as a public benefit corporation, most of our work in pursuit of our mission takes place inside the company. Research relevant to AI safety and interpretability, Applied biomedical research. Discounts on Claude for nonprofits and scientific institutions. These activities have both a public benefit component and for-profit value, so we house them directly in the company. The Anthropic Public Benefit Fund will have increased flexibility and scale, enabling us to pursue our mission in new ways. For example, we might explore large donations to traditional nonprofits or direct distributions to specific populations. The example they give is labor market impacts, where they say not only might they use this money to study these issues, But they imply heavily that they might just give money to people as well. As they put it, given the pace of AI progress, bolder approaches may be needed. They do note that this is not going to be a 501, the traditional structure for US nonprofits. Instead, it will be a wholly owned LLC, which they say is to, quote, give adjustability between different forms of giving. They do pledge that none of the equity held by the LLC will be used for lobbying or political giving, although many of the skeptics are specifically skeptical about that. A number of Anthropic employees took the chance to say that this was part of what made them passionate about their company. Engineer Jason Clinton said that he had pledged half of his equity and added, this was the principal reason that I left Google for Anthropic. The existence of this kind of program was a costly signal that Anthropic wasn't just talk at a time when Anthropic didn't even have a released model. Fund a bright future. Now, what's interesting is that although we are starting to get details, The philanthropic world has been looking over at AI wealth for a while and starting to prepare for a very different paradigm. Earlier this year, Nan Ransohoff, a Stripe executive focused on public good, wrote a blog post called The Third Wave of American Philanthropy. Between the OpenAI Foundation holding 26% of OpenAI and Anthropic's 7 co-founders pledging to give away 80% of their wealth, Nan wrote, I was doing some simple napkin math to wrap my head around the scale of what's coming and radicalized myself in the process. I had dramatically underappreciated the scale of the philanthropic capital that's about to become available and the corresponding gap in talent and organizations that will be needed to make the most of it. Nan estimates that over the medium term, AI will likely add between $37 and $100 billion per year in intended philanthropic spend. Now, that's estimating the OpenAI Foundation's 26% at the company's valuation at the time of $850 billion and Anthropic employees' and co-founders' commitments at their then-current $900 billion valuation. Obviously, those numbers are all smaller than what we're dealing with today and what we anticipate at the IPO. Added up, Nan estimates $370 billion in total philanthropic assets. The $37 billion she gets to by guessing at everyone wanting to spend around 10% per year. The 10% she gets to by taking the in-between on foundations that pay out an average of 5 to 9% of their assets per year versus donor-advised funds that pay out an average of 20 to 25%. She points out that that $37 billion per year could go up quite quickly. If A, the valuations double, she writes, in a handful of years, although that's basically already where we are, and if, as she puts it, maybe these donors want to increase their target annual spend from 10 to 15% because they're especially worried about the AI transition. That's what gets her to $100 billion a year in target annual spend on the top side. Compare that to current US charitable giving, which is around $600 billion per year. This would represent somewhere between a 6 and 17% increase in annual spend. Now, it's a little bit beyond the scope of today's show, although if this is something that's interesting to you, I will absolutely get into it, given that a lot of my early background is in the philanthropic and social entrepreneurship space. But the rest of Nan's post gets into why it's actually going to be quite difficult to positively spend all that new money. She points out that, in her words, we're short thousands of philanthropic startups and the talent to run them, that we need elite capital allocators who can deploy $50 billion a year well, and that neither of these things is super easy to fix. Like I said, there is a lot of thinking that's going into this right now. So if this is a subject that's of interest to you, I'm happy to go deeper. Coming back, however, to the Anthropic Public Benefit Fund, there were, as you might imagine, a lot of responses. On the negative side, a lot of the folks that view Anthropic as the chief purveyors of the doom and gloom narrative are concerned about all the funding going to that. Writes accelerationist spiritual leader Beth Jezos, this is going to be doomer NGO bonanza forever. Given Anthropic's tussles with the administration this year, there are also plenty of right or right-leaning commentators that reject the idea that this money won't be political. Summing up that strand of discourse is Polymath from X who writes, this is going to be an NGO slush fund. This is a bribe to left-wing politicians to get them on board with AI so that they will follow Anthropic's lead on any policy or regulation. Perhaps the more, if not nuanced critique, then at least open question around this was summed up by former AI czar David Sacks, who said, you've put 20% of your equity in an LLC. Whether it serves the public good remains to be seen. And there are plenty of folks who are asking in this context, what does public good or public benefit actually mean? Pointing to the part of Claude's constitution where Anthropic discusses principles encouraging consideration of non-Western perspectives, Anduril CEO Palmer Luckey writes, the problem is that their definition of public good is helping everyone else, not the American public. If they would commit to helping America and Americans, that would be one thing. But this reeks of USAID 2.0—just another way to funnel billions out of our own economy. Bloomberg's Joe Weisenthal, on the other hand, thinks it's fairly rich for arch-capitalist entrepreneurs and VCs to be telling Anthropic what to do with their funds. He posts lots of people telling another company how they should spend their money. But in some ways, this is just speedrunning and accelerating. difficult conversations that have always been had around philanthropy. Jeremy Flores on X posts, the public good is something determined by the public itself, and the public has a say in its administration via democratic controls over a democratic state. This instead is your company stealing everyone's work, dismantling the state, and then deciding who gets resources. Now hold aside the very clear position that Jeremy is coming from, given that he's accusing Anthropic of stealing everyone's work. The first part of his post actually gets to a question of philanthropy, that has been around since the Gilded Age. Social reformer Jane Addams wrote a piece for The Atlantic in 1899 called The Subtle Problems of Charity. This was, of course, another moment in American history where people were very uncomfortable with the recent power acquisition of private enterprise relative to other institutions like democratic institutions. And the essay's a meditation on how far philanthropy can go and where democratic process has to begin. Addams wrote, Probably there is no relation in life which our democracy is changing more rapidly than the charitable relation— that relation which obtains between benefactor and beneficiary. At the same time, there is no point of contact in our modern experience which reveals more clearly the lack of that equality which democracy implies. We find in ourselves the longing for a wider union than that of family or class, and we say that we have come to include all men in our hopes, but we fail to realize that all men are hoping and are part of the same movement of which we are a part. Many of the difficulties in philanthropy come from an unconscious division of the world into the philanthropists and those to be helped. It is an assumption of 2 classes, and against this class assumption our democratic training revolts as soon as we begin to act upon it. And so bringing it back to the Anthropic Public Benefit Fund, what would it even look like to give the beneficiaries more control in how the resources are used? Lawyer Gabriel Shapiro wrote, notice how they always have to stay in control. God forbid you just airdrop 5% on the public. No, they have to be able to monetize their own stock while determining how to spend it on things deemed by them as benefits. Perhaps a more impactful version of that comes from Elon Musk, who said, I recommend giving the shares directly to the public. This is what I will be doing. Andrew Curran pointed out that this actually reminded him of another company's policy. He wrote, Elon says that instead of putting shares into a public benefit fund, he will be giving shares directly to the public. Back in April, OpenAI also proposed a public wealth fund that would distribute returns directly to U.S. citizens through a dividend. Now, as to the mechanism of this, Mach 33 founder Aaron Burnett wrote, if Elon means by donating via Invest America, i.e., Trump accounts, then I think this is a very good idea. The single most optimistic future is one where the entire next generation owns a piece of SpaceX. Ownership is amongst the most American ideals. Owning a piece of a spacefaring future is next-level American optimism. Elon simply responded to that, yes. For some, the interesting question wasn't about the specifics of the redistribution, but about what was going on inside Anthropic that made it feel like this was important to discuss now. In short, Fable-55 is the best model in the world, and we're so confident it'll print money that we're planning how to redistribute the wealth of a model that doesn't even exist publicly yet. What on earth do Anthropic have internally to be doing this? Commenting on a similar post, 4lex4 referenced another recent Anthropic research paper saying they've seen 17% GDP growth in 2030. Now, the report that had that 17% number also had a detail about Anthropic reducing and then ultimately ceasing to hire junior technical talent in fairly short order. Joel Stein wrote, pretty wild that we're seeing this about the Public Benefit Fund at the same time we're learning that Anthropic is on track to cease hiring junior technical talent by mid-2027. And indeed, to some, this is basically Anthropic announcing privatized UBI. AI aggregator Andrew Curran again writes, the most significant line for me is direct distributions to specific populations. It sounds to me like they are planning their own form of self-directed global universal basic income. Now, getting at why a private UBI is quite different than a government program, Prinz on X writes, you can't have UBI that is also only to specific populations. Now, one group who was interestingly absent from the discourse were the AI safety folks. And looking around, it seems like they were too busy being angry at Anthropic about their new proposed AI safety bill to comment on the Public Benefit Fund. Politico reports that Anthropic is pitching their own AI safety bill to lawmakers right now. A confidential draft has been quietly circulating around Washington and was provided to Politico last week through an anonymous source. An Anthropic spokesperson confirmed the existence of the bill and commented, This bill represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June. These are sensible policies we believe Congress should act on. The bill would allow the government to block the release of a model if it poses a foreseeable imminent risk of causing more than $1 billion in damages or killing more than 50 people. The risk must arise from the creation of a bioweapon, an AI model evading the developer's control, or a cyberattack. This is similar to the way safety thresholds are defined in state AI bills in California and New York. In addition, the bill would require independent evaluators to assess safety protocols at Frontier Labs. These evaluators would be required to register with the Commerce Department, but would be independent of the government. The evaluators are tasked with reporting imminent enumerated risk to the Commerce Department, establishing the basis to block a model. Blocking a model's release or use is adjudicated through the federal court, rather than as an administrative matter within the Commerce Department. And notably, the language of the bill expressly prohibits any other department, agency, or officer from suspending a model's development, deployment, or internal use. That seems to be targeted at similar incidents to the Pentagon's use of the supply chain risk designation to disrupt Anthropic models from being used by military contractors. Finally, the bill includes various safety requirements from the state legislation, such as mandating the creation of safety protocols and reporting of major safety incidents. As I said, the AI safety folks are not happy about this. If anyone builds it, everyone kills it, co-author Nate Suarez writes, Anthropic's proposed law would make it even harder than it already is for a court to block or restrict the development of an AI that threatens all humanity. Others were mad at things like federal preemption from state rules. Jeffrey Ladish writes, this seems insanely dangerous coming from Anthropic. Dario has called on the government to slow things down, and this bill would remove the ability for states to do exactly that while making it very difficult for the federal government to actually stop dangerous training runs. Now, others pointed out the inherent problem with definitions in these types of bills. David Sacks pointed to the problems of the definition of frontier. He wrote, a new Anthropic-written safety bill is circulating on Capitol Hill. Some of its ideas could be reasonable if limited to true frontier companies. As written, however, it will eventually sweep in most of Silicon Valley, damaging the startup ecosystem. The bill defines the frontier as any model trained on 10^25 FLOPS. The EU standard from 2023! Biden's executive order, California's SB 53, and New York's RAISE Act use 10^26— 10 times more. 10^27 is where the frontier actually is today. Astra was reportedly trained on 100,000 Grace Blackwell GPUs. A 10^25 model can be trained on as few as several hundred GB300 chips in 3 months. That is not frontier in 2026. The frontier keeps moving. Any definition meant to track risks should move with it. Now, on the totally opposite end of the political spectrum, New York politician Alex Borras, who was targeted very intensely by pro-AI PACs earlier this year, wrote, Anthropic is pitching a bill that closely matches its framework from June. How can you watch Hugging Face, Coxin, and rogue AI swarms and conclude that the bill you cooked up in June is sufficient? This comes nowhere close to the speed of developments, nor the speed we need to keep people safe. What I think is interesting is that both Boras and Sachs are, from very different perspectives, saying that this bill is basically outdated, although in very different ways. For many folks, the subtextual reading of this is Anthropic actually trying to limit the powers of the executive when it comes to AI risk. Adam Connor wrote, this bill could most charitably be described as the best way for Anthropic to defend itself from abuse from the federal executive branch. by limiting their power to take action and taking away everyone's ability to do so. Journalist Kelsey Piper expanded. She wrote, one unfortunate conclusion I'm reaching from this is that Anthropic, while acknowledging they're doing something extremely dangerous, is more importantly with regard to policy terrified of the executive branch having the ability to flexibly make them stop at the drop of a hat. This isn't shocking given that they've been on the receiving end of the executive branch doing that in an unprincipled and stupid way. Anyone who has dealt with a bureaucracy becomes a libertarian, And the administration's treatment of Anthropic was genuinely outrageous. But it means that they're interested in giving themselves more process-based protections when everyone else is interested in giving us more process-based protections against them. Here's where this nets out for now. What's coming up in the episode today are 2 massive issues related to AI that appear independent but become linked very quickly. One is the AI wealth problem, the immense success of these companies is concentrating an extraordinary amount of wealth in a relatively small number of hands. And the public benefit fund in some ways amounts to a recognition that some amount of that wealth is going to have to be redistributed some way. It's either going to be redistributed voluntarily, or it's going to be redistributed by someone doing the redistribution. And Anthropic clearly, and perhaps understandably, wants to be doing that redistributing themselves. And yet this is running headlong into another problem, which is concentration of power. And a growing concern that owning these models and the wealth that comes with them involves too much power for one private institution. All the discourse and debate around the fund and around the safety bill are really all proxies in some ways for the larger question of how AI wealth and power is going to interact with democracy. Those are not easy questions, but they're certainly ones we should all be involved in. It is a subject that I anticipate growing in people's focus and recognition, but for now, we're gonna leave it there for today. Appreciate you listening or watching as always, and until next time, peace. Peace.