← The AI Daily Brief: Artificial Intelligence News and Analysis
The AI Daily Brief: Artificial Intelligence News and Analysis
Why Fable 5 Is the Most Controversial AI Release Ever
00:30:25
Loading summary
Transcript1 lines
- [00:01]
A
Today on the AI Daily Brief why Fable 5 is easily the most controversial AI model launch of all time. Before that in the headlines, more chatter about the AI labs donating equity to a sovereign wealth fund. 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. From first of all, thank you to today's sponsors, KPMG, Robots and Pencils, ZenCoder and OutSystems. To get an ad free version of the show, go to patreon.com aidaily brief or you can subscribe on Apple Podcasts. To learn more about sponsoring the show, send us a note at sponsorsidailybrief AI now speaking of aidailybrief AI, this has come up a couple of times in the past few days in and around my discussions of Fable, but given the increase in capability to just get things done, I have officially pulled the trigger and launched a new AI Daily Brief website. This one is meant to answer the most common request that I get, which is make it easier for us to share specific parts of the episode. Each episode then is going to have a summary page that includes the one big idea, the key numbers from throughout the episode, and 15 to 20 shareable cards that have individual insights, individual quotes, along with timestamps and the ability to link those specific cards out for folks who want their agents to grab it. You can also download everything on this page as markdown or get the official episode transcript. Now. I decided to push this out fast rather than have it perfect, so there's going to be a lot of rough edges. Please shoot me an email if you have any feature requests and I hope you enjoy the new AI Daily Brief. AI we kick off today with President Trump once again calling for a sovereign wealth fund seeded by AI Equity. In a press conference in the Oval Office on Wednesday, Trump said, I'm going to have meetings with the top 12 or 15 executives very shortly and we're talking about giving something back to the public. And if we do that, the public will become very rich, the people in our country, because that's the kind of money we're talking about. And I think they'll do that and I think it will make it very popular now. The New York Times noted that Trump's comments have, quote, turned up the temperature on a hot topic in Washington and Silicon Valley as the tech industry reckons with a growing backlash against AI. Sources said that Sam Altman did not discuss the idea directly with Trump during his visit to the White House last week. However, the concept of a sovereign wealth fund was heavily discussed in Altman's meeting with Bernie Sanders. In good news that the world has not gone completely topsy turvy, Altman did reportedly object to Sanders proposal of OpenAI giving 50% of their equity to the public. Back in Silicon Valley, Altimeter Capital founder Brad Gertzner is warning the AI companies might need to pay some form of anti revolutionary tax. During a conference panel this week, he said it is destabilizing when you're creating trillions of dollars in private value and 80% of Americans think it's a scam where they're left out and left behind. Now, while there is some amount of groundswell around the idea, White House officials said discussions are in very early stages, stating that they're unaware of any real plan or vehicle to acquire stock. Summing things up, David Yaffe, a professor at Harvard Business School, said the notion that the government should be a partner in new technology is not new, but the idea that the government and American citizens should have equity is a radical departure from the free market's approach of the world. Then again, the New York Times suggested that all of this might not be some grand plan to reshape the U.S. economic system for the AI age. It's instead arguing that the concept of an AI public wealth fund appears to have one central animating idea. As they put it, the president likes owning equity in businesses next up, whatever the ultimate state of their equity looks like. OpenAI is closing a deal to stand up 10 gigawatts of data center capacity Sources told the Information that OpenAI is in advance negotiations to lease a 10 gigawatt data center campus to be built on federal land in Ohio. The campus is expected to cost $500 billion based on current pricing for chips, labor and power. This would make it not only OpenAI's largest data center commitment to date, but likely the largest data center campus ever built. 10 gigawatts is roughly four and a half times the total output of the Hoover Dam and more than twice the capacity of the largest nuclear power plants operating in the U.S. sources said that Nvidia is attached as a financial backer, and the current structure of the deal would see OpenAI lease the chips as well as the facilities. They wouldn't be required to begin repayments until the first GPUs are powered up. With 800 megawatts expected to come online in 2028, the total life of the lease could be as long as 20 years. And while one of the critiques of Nvidia in the past has been participating in some sort of circular financing deals, they haven't previously backstopped any data center deals whatsoever, let alone one of this size. Instead, the deal has echoes of Project Stargate, which was announced at the beginning of 2025 but stalled out earlier this year. Stargate was originally intended to be a $500 billion joint venture with Oracle and SoftBank, who appear to be playing a role in this new deal as well. The site in question is a decommissioned uranium enrichment facility in Pike County, Ohio, around 50 miles south of Columbus. The federal government will own the power plant, with SoftBank subsidiary SB Energy functioning as the plant operator now. While it appears that the project is being designed to preempt the major sources of data center outrage, local politics are still turning sour. Controversy erupted this week in Cleveland when legislators discovered their previous governor had signed deals with Amazon, Meta and Google to provide 100% sales tax exemptions on data center operation lasting for 40 years. While the deal made Ohio an attractive destination for new data centers and has already brought in tens of billions in investment, they're the state estimates the deals will cost 1.8 billion in lost tax revenue and as one report states, given the uncapped nature of the tax exemption, final numbers could be significantly higher, democrat Representative Tristin Radar told Local Press, I'm just dumbfounded this is so poorly designed. The tech companies saw fertile ground and have been taking advantage of us ever since. Meanwhile, overall bans and moratoriums are increasingly on the table as data center resistance grows. New York looks set to become the first state to hit pause on new data center construction after passing a one year moratorium late last week. The bill would block any new permits for data centers above 20 megawatts, which is obviously very small scale for AI infrastructure. State Senator Kristen Gonzalez, the lead sponsor of the bill, argued that New York's constrained and aging grid can't keep up with the pace of construction. Almost 10 gigawatts of new facilities are currently seeking approval. Next up, we wait to see whether New York Governor Kathy Hochul will actually sign this one. The City of Seattle followed suit on Tuesday, with City Council unanimously approving a one year ban on new data centers as well. The proposal came after the Seattle Times reported that five proposed data centers could consume up to a third of the city's electricity. Interestingly, the opposition in this case was driven by tech workers themselves, with groups including Amazon Employees for Climate justice mounting a letter writing campaign. The reasoning was a little garbled, with a spokesperson for Climate Group 350 Seattle claiming tech workers organized against the data centers because, quote, AI is synonymous with people losing their jobs. The antipathy towards data centers has even reached Texas, with Governor Greg Abbott calling for stronger consumer protections against rising electricity costs. Abbott directed state utilities to require new data centers to fully fund additional infrastructure needed for their operations to ensure the costs aren't pass ratepayers. In addition, Abbott unveiled a wide ranging regulatory agenda for consideration in next year's state legislative session, including mandatory closed loop cooling systems, annual reporting of water and electricity use, and a requirement for data centers to add new electricity generation to the grid. Now to be honest, if you are a person who thinks on the one hand the single fastest way towards guaranteed AI inequality is data center moratoriums, but on the other hand completely recognizes communities not wanting to foot the bill for increased electricity costs along with dealing with other type of negative externalities of data centers, then looking to a place like Texas, which is a very popular destination for new data center construction, to lead the way in setting standards for what that construction should look like is kind of optimistic. Certainly the money for data continues to flow Broadcom is getting in the data center financing game with a $35 billion fund backed by Blackstone and Apollo. The initial 35 billion will be used to fund 1 gigawatt of capacity deployed at sites operated by Fluidstack using Broadcom's custom AI chips and networking infrastructure. The first project will be delivered to Anthropic. The fund is intended to provide a stable multi year capital structure to facilitate the long term compute buildout. The partnership is aiming to fund 20 gigawatts of capacity for leading AI labs through 2028. Apollo and Blackstone will serve as capital providers syndicating the fund out to their investors. Broadcom and Anthropic are world class companies operating at the frontier of technological innovation and we are proud to have led the largest private financing ever executed. AI compute is rapidly emerging as one of the most compelling new asset classes in finance, characterized by contracted cash flows, mission critical utility and a supply demand dynamic that continues to intensify. Won Kim, the head of corporate development and infrastructure partnerships at Broadcom, commented, the demand for AI COMPUTE is growing faster than traditional capital markets can accommodate and this initial transaction led by Apollo demonstrates what becomes possible when world class technology is paired with a partner of that caliber. Lastly today, despite revenue figures remaining strong, huge cost overruns and an ambitious fundraising plan overshadowed Oracle's solid earnings during their Wednesday night earnings call. Oracle reported 16.5 billion in capex spend for the previous quarter which brings their annual total to 55.7 billion, which was above their $50 billion forecast. Oracle said they plan to raise spending to 70 billion for the coming fiscal year, but reported costs will come in between 20 and 25 billion higher due to prepayment of some components to deal with rising costs. Oracle plans to raise another 40 billion in equity in debt in the coming fiscal year. Oracle raised 48 billion in debt in equity over the past fiscal year and now carries 117 billion in total debt. Revenue figures remain strong with 21% growth to reach 19.2 billion for the quarter. Cloud infrastructure sales are growing at a 93% pace, meanwhile. Still, investors couldn't look past the pile of mounting debt and the stock plummeted 11% in after hours trading. Now, this is not primarily a market show, but there is some interesting new discourse making its way to Wall street. Stuff that frankly, listeners of this show would already have been hearing about, and it's likely we'll dive a little deeper into that in the coming days. For now, that is going to do it for today's headlines. Next up, the main episode. One of the most important AI questions right now isn't who's using AI? It's who's using it? Well, KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising the highest impact users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more@kpmg.com US sophisticated that's kpmg.com US sophisticated Here's a harsh truth. Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized. Half of companies have AI tools, but only 12% use them for business value. Most employees are still using AI to summarize Meeting Notes if you're the one responsible for AI adoption at your company, you need section Section is a platform that helps you manage AI transformation across your entire organization. It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value the result? You go from rolling out tools to driving measurable AI value. Your employees move from meeting summaries to solving actual business problems, and you can prove the ROI. Stop guessing if your AI investment is working. Check out section@sectionai.com that's S E-C-T-I-O-N a I.com so coding agents are basically solved at this point. They're incredible at writing code. But here's the thing nobody talks about. Coding is maybe a quarter of an engineer's actual day. The rest is standups, stakeholder updates, meeting, prep, chasing context across six different tools. And it's not just engineers. Sales spends more time assembling proposals than selling. Finance is manually chasing subscription requests. Marketing finds out what shipped two weeks after it merged, ZenCoder just launched ZenFlow work. It takes their orchestration engine, the same one already powering coding agents, and connects it to your daily tools. Jira Gmail, Google Docs, Linear Calendar notion. It runs goal driven workflows that actually finish your standup Brief is written before you sit down. Review cycle coming up. It pulls six months of tickets and writes the prep doc. Now you might be thinking, didn't openclaw try to do this? It did, but it has come with a whole host of security and functional issues which can take a huge amount of time to resolve. Zencoder took a different approach. SOC 2 type 2 certified curated integrations, tighter security perimeter, enterprise grade from day one, model agnostic and works from Slack or Telegram. Try it at ZenFlow free. This episode of the AI Daily Brief is brought to you by Outsystems, a leading agentic systems platform built for the enterprise. Organizations all over the world are building, orchestrating and governing agentic systems on the Outsystems platform and with good reason. Outsystems open and unified platform allows teams to architect, deliver and scale governed agentix systems. With agility, teams of any size and technical depth can use Outsystems to build, deploy and manage AI apps and agents quickly and cost effectively without compromising reliability and security. With Outsystems, you can rapidly launch ideas from concept to completion. It's the leading agentic systems platform that is unified, agile and enterprise proven, allowing you to accelerate growth, reduce operational friction and deliver real enterprise impact with AI Outsystems build your agentic future. Welcome back to the AI Daily Brief. There is apparently something cursed when a big AI lab launches the first 5 edition of their big model. Last August it was GPT5 which caused intense controversy around the choice to deprecate the 4.0 model. Not to say other questions around the performance of GPT5, but the sheer tonnage of the backlash to Fable 5 makes that look like nothing. Indeed, it took less than 24 hours of intense response for Anthropic to walk back policy changes, telling Wired we made the wrong trade off and we apologize for not getting the balance right. But how did we get here? Why was there so much anger? And why do I think that this particular issue is coming to some sort of resolution? It portends much bigger fights ahead. Summing up the mood late in the afternoon yesterday was Matt Palmer, who wrote Anthropic People, you've got a couple of days at most to mitigate the damage being done by your senior leadership and policy people. The stealth nerfing and data retention decisions are titanic screw ups that pose serious risks to both your technical pole position and your bags. And honestly, it wasn't even just these things that people were complaining about. First of all, as we discussed on yesterday's show, there were some pretty strict safeguards put into place around things like biology. Now I will fully admit that on yesterday's episode I lamented the fact that a lot of the initial response to this I saw seemed to me to be coming from people who were just looking for something to gripe about. And yet, as the course of the day went on, there were more and more stories showing just how overwired these guardrails were. Biomedical engineer Daria Anutmaz, who is, by the way, always given early access to models and who is a huge booster of all of the labs, tweeted, I can't even say hello to Fable 5 except in incognito mode. That is memories off because it knows I'm a biomedical researcher. It would be nice not to ban biomedical scientists, but I think even if it were that alone, it could have been fairly easy to come to some sort of resolution over a period of time. The problem was that underneath that, the revelation of the power to determine who gets access to the tools of the new economy is something that people are getting more and more unsettled with now. Another issue was the data retention policy that we discussed, where basically enterprises would have to be okay with anthropic keeping their messages, even ones that they deleted, for 30 days. Lawyer and AI user Prinz writes, The 30 day retention policy is not fine at all. First, it applies only to zero data retention customers, that is those customers that have specific reasons to ensure that anthropic can't see their data at all. Second, anthropic employees get to see both prompts and outputs quote flagged for potential serious harm or at a customer's written request. What is flagged for potential serious harm? If I'm a lawyer representing the government in a case involving allegations of improperly conducted surveillance. Is that potential serious harm? What's a customer's written request? Which customers are we talking about? The NSA is a customer Anthropic. If my law firm were using Claude, I would tell it to lock us out of Mythos and Fable immediately and transition us to another provider, he continued in another tweet. Reasons of safety are all well and fine, but enterprises will now find themselves in the same position as the DoD. Anthropic is asserting its right to look at their private communications with CLAUDE that have, quote, potential for serious harm. A vague phrase defined by Anthropic in its sole discretion. I am extremely uncomfortable with this notion, by the way. Sympathy for the DoD was one fairly widespread and pretty unexpected response to Fable 5. Investor Yonda Ehrlich writes, I feel like we all got the same treatment from Anthropic that the Dow got in February. I viscerally understand the Dow reaction now. Now what's more, this data retention issue wasn't just theoretical. It took about one hour before the Verge reported that Microsoft had started restricting employees from using Fable 5 and Copilot because of those data retention concerns. Matt Palmer again wrote, cannot think of a more disastrous set of decisions to make ahead of an ipo. The reaction to data policies alone will show up in their revenue figures, to say nothing of cost control measures. Okay, so we've got two things now. We've got the safety classifiers in general. Which arena AIs Peter Gostev may have had the best response to writing A year ago or so I participated in Anthropic safety testing program trying to get past their safety classifiers. I remember thinking this is stupid. These trigger on everything. Nobody would ship a model like this to production. Turns out I was wrong. So we've got that. Plus we've got these data retention policies which we're already starting to see enterprises respond to. But the biggest thing, and the thing that really got everyone fired up was the new limitations on how LLMs could be used for other LLM development. The relevant part of the system card was in light of the ability of recent models to accelerate their own development, we've implemented new interventions that limit claude's effectiveness for requests targeting Frontier LLM development, for example on building pre training pipelines, distributed training infrastructure or ML accelerator design. Using CLAUDE to develop competing models already violates our terms of service, but enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms. And yet for many, the worst part was in the next paragraph when they write Unlike our interventions for cybersecurity biology and chemistry and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter efficient fine tuning. In other words, this was to be a silent nerfing of the model. Akashgupta summed up Anthropic is now silently making Claude Dumber for certain users on purpose, and there is no way to tell when it's happening to you. If your request looks like frontier LLM development, Fable 5 degrades its own output through prompt modification, steering vectors, or parameter efficient fine tuning. You'll never see a refusal, and you never get switched to a weaker model. The answers just get worse. Think about what that breaks benchmarks. Assume the model you tested is the model you get. That assumption just died for an entire category of work. An ML engineer debugging a failing training run can no longer distinguish the model is wrong from the model was made wrong on purpose and classifiers misfire. Samyanalysis says GPU inference research is already getting caught, and inference optimization is what every company running open models does, not just Frontier Labs. A false positive on a visible refusal is annoying. A false positive on silent degradation is undetectable, he concludes. The strategic logic is real. Anthropic has said models now accelerate their own development, which means that they know precisely how much a frontier model compresses Frontier timelines. Renting that compression to competitors for $200 a month is a genuine cost. But the precedent is the story. Once silent degradation ships as a feature, every eval needs an asterisk. Results valid unless the lab decided your use case shouldn't work, research group Alpha XIV writes. As believers of open research, we are disappointed to see anthropics silently degrading Fable 5 for AI development. Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing, this sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider. This is not safety. Safety policies should be transparent, auditable, and user visible. On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It's the independent researchers, academic groups, startups and open source builders who rely on public tools to compete, innovate and pioneer AI. For everyone else, Rohit on X writes, the issue isn't the existence of safeguards, it is that the classifier is terrible, exceedingly trigger happy, unusably so for many that it silently degrades responses. If it's about AI and that it captures all user data. These are actively bad, not just a mistake. There are real trade offs in Safety, and this release chose none of those. It basically nerfed the model in the most blatant way possible, taking none of the nuances into account. AI safety researchers can't use it, bio researchers can't use it. Cybersec researchers can't use it. Even by the system Card's own admission, this isn't in immediately developed superweapons territory, which makes it even more egregious. They did it because they can, which invites scrutiny. How can you trust Anthropic to do the right thing when it counts? We just had this argument about their fight with the Dow that Anthropic didn't want to be the final arbiter, they just wanted safety. This is the opposite. They really do want to be the final arbiter. David Shapiro agrees, arguing in different words that this is Anthropic wanting to be final arbiter because of where it stands vis a vis AI safety. He writes, to anyone who doesn't understand why Anthropic is silently nerfing AI and ML research specifically, it's because they believe in Yudkowskian delusion of fum, whereby achieving recursive self improvement means that the extinction of humanity is nigh. Thus they believe that hamstringing their one model will meaningfully slow down AI development and research fellow Tom Davidson, who, to be clear, ultimately came to the conclusion that silent sabotage is a scary precedent and the wrong call gave the Steelman argument for Anthropic's decision to secretly nerf AI R and D. And it was about exactly what the critics said it was. He argued that the strongest defense of Anthropic was that one by far the biggest risks are from super intelligent AI. Two to manage these risks, the leading company will need to pause partway through the intelligence explosion because pausing allows them to generate evidence of misalignment that would be needed to justify a longer global pause, as well as to use powerful AI to accelerate alignment progress. That 3A pause is much more likely if the leading company has a big lead and much less likely if multiple companies are neck and neck that 4 if lagging AI companies can use the leader's AI for AI R&D during an intelligence explosion, the leader cannot maintain their lead. And so five sharing AI R&D access with competitors massively decreases the chance of a pause at the critical time and massively increases the risk from superintelligent AI. And six and this is a big one, Anthropic can't block competitors using Mythos without the silent sabotage. It's very hard, tom writes, for a frozen safeguard to block someone that can iterate against it. It sucks that this is the only way, but it is. And this idea, even if it's not fully the way that Anthropic is looking at things, has the feel of being close enough that many people are for the first time grappling with just how much power the leading AI lab or labs are going to have and they don't like it. Imbue CTO Josh Albrecht writes, why stop there if Anthropic was serious about safety? They shouldn't just degrade output quality, they should insert backdoors, exfiltrate your data, ban your account, and brick your computer. Wouldn't want to increase existential risk by letting random humans do science. CMU's Graham Newbig writes, first they came for the model builders. I feel like we're getting a glimpse of the future where AI is only provided to a privileged few, and that's not a future I want to live in. Anthropic's communications did not make it any easier on themselves. Two things were flying around that I think very inarguably made everything worse. The first was that Anthropic CEO Dario Amadei wrote a typically massively long piece called Policy on the AI Exponential that, in the context of the discussion that we were having, did not sit well with many con. Grogan summed up what many readers felt when he said TLDR1 declare AI too dangerous for ordinary competition, so you propose a regulatory regime where only the largest incumbents can survive. 2. Warn about labor displacement while selling the product to executives as a Labor Displacement Tool. 3. Warn about state overreach while asking the state to license and gatekeep Frontier models. 4. Warn about corporate power while sketching a corporate state cartel over compute release, security, export controls and deployment. So that was one problematic document. And then the second was Bloomberg originals released this 47 minute deep dive on the company that, to put it lightly, did not do anything to decrease people's concerns around the role that Anthropic appear to be putting themselves in vis a vis the world. Powerbottom dad, who is a market commentator outside of the AI space, quoted Dario in the video when he said I'm scared of government having it and added, yeah, this guy thinks he should be God emperor of A.I. sole decision maker. Now the point here is of course not that everyone agrees with that analysis, but that when you take the combination of the communications coming out of Anthropic, the decisions they are making around policy, and the inherent power of their position, it makes this take feel a lot more reasonable than it might once have. Finbaronx wrote, as my entire feed is criticizing Anthropic, I think that the team there genuinely believes what they're saying. It's not a marketing or anti competitive tactic. They genuinely believe these models are dangerous and that AI research should be slowed down. GMU Law professor Samuel Roman retweeted that, however, and added, I think this is the true and fair reading of Anthropic's actions, but the real problem is that it reveals a real level of hubris vis a vis other societal actors. The only way their decisions make logical sense is if they presume that they will maintain control over the frontier to dole out access to it without pushback from those other actors. Regardless of whether you believe that this disposition is morally justifiable, there's a practical consideration about maintaining control of it I don't think has been thought through. If Anthropic genuinely establishes itself as the tollbooth for frontier model access, the state is absolutely going to correctly read that as a direct form of competition and act accordingly. And I'm sorry, but Anthropic just does not win that fight. This means the future direction of models just became much more likely to be determined by a small group of bureaucrats via edict rather than a broad societal discussion and development. I personally trust the latter much more than the former, hence the disappointment. Everyone I've met at Anthropic has been wonderful and well intentioned, but there's a degree of myopic focus on the mission that I think often ends up blinding them to the fact that these are not single player games. Mostly there was just a lot of frustration. Nekonx writes. The actual audacity Anthropic absorbed the collective intelligence of the world and is now trying to create a class system where that very intelligence is gatekept from anyone and everyone who they deem is not worthy. Bubble Boy writes, everybody is scared of Chinese models because it won't let you criticize the ccp, while Anthropic won't let me use their models for life saving medical research. Who's the real villain again? Summing it up, Bazelord writes, if future human actions will be largely taken via AI models, we need to seriously consider what it means for AI labs to arbitrarily control what people are and aren't allowed to do. And this is what I mean when I say hold aside the specific decisions that Anthropic is making. Hold aside the specific communications that may or may not make those decisions worse. There is an inherent power in their position that people are finally coming around to that potentially gives them far more power over people than any private corporation has ever had. Now back to the change in specific policy. Like I said, it took less than 24 hours for anthropic to walk back the policy around nerfing LLM research. In a statement to Wired, Anthropic said, We're changing Fable 5 safeguards for frontier LLM development to make them visible. We made the wrong trade off and we apologize for not getting the balance right. Basically. Now Fable safeguards for AI development will be visible. As Wired puts it, if the company suspects a user is trying to use Claude to build a highly capable AI, it will alert them that it's either refusing the request or rerouting the user to a less capable model. AI policy expert Dean Ball writes, this resolves the central concern I had with the Fable release, which was the silent degradation. I'm glad to see Anthropic make the right call here. That said, I suspect the residual broken trust and resentment this has created will linger and have a blast radius wider than Anthropic hugging faces. Arthur Zucker writes, dear Anthropic, you broke our trust and I don't think you'll ever get it back. My tokens will no longer fly your way. David Kramer makes another good point. Even if you don't have an ethical or moral stance on the limitations of Anthropic's models, there's a practical business one I will avoid Anthropic models because they keep imposing more limits on the products I can build. I'm not going to build on a completely walled off ecosystem. Now as we wrap one part of this story is about the implications for Anthropic Factory CEO Matan Grinberg writes, Anthropic speedrun to becoming the bad guy should be studied and truly from an unassailable position just a few months ago to where they are now is quite wild for what my advice is worth, now that this particular LLM research policy is resolved, I would be looking very quickly into that enterprise data retention policy, because there are a heck of a lot of users, the users in fact, who have made Anthropic the juggernaut that it is, who cynically might not care at all about the LLM research question but you better believe aren't going to stick around if their corporate data is subject to Anthropic's whimsical When it comes to competition and next moves, the ball is clearly in OpenAI's court, although it's not clear exactly what step they'll take. The information reported on a Sam Altman Slack message that many people took as indicating that their next AI model, release 5, 6, wasn't at current state up to fable standards. The Wall Street Journal, meanwhile, last night reported that OpenAI is considering significant price cuts on their tokens, which could create a pricing war with fairly significant impacts for this whole industry. Although at this point it's just a report and clearly just one of many things being discussed. Ultimately, if you are looking for the positive side of this, it's a whole lot of folks waking up to the stakes of the role of the labs in society hugging faces. Clem Delang wrote, Concentration of power capabilities and economic wealth is the biggest risk in AI. We need open science and open source more than ever. I think the fallout from everything that has happened right now is just like the fallout after GPT5, going to cast a long shadow on the next stage of development, and I will of course be here to follow along as it happens. For now, however, that's going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.