Path Toward Resolution and Future

Today in the AI Daily Brief as the workwk starts without resolution, the Fable five crisis continues. The AI Daily Brief is a daily podcast and video The most important news and discussions in AI friends, quick announcements before we dive in First of all, thank you to today's sponsors, KPMG, Section, Assembly and Out Systems To get an ad free version of the show go to Patreon dot com slash AI Daily brief, or you can subscribe at Apple podcasts And if you want to learn more about sponsoring the show, setnd us a note at sponsors at AIDailybrief. ai Today we are picking up where we left off in the chaotic saga of Anthropic versus the White House around Fable fiveive. For those just getting up to speed, on Friday night, the United States issued an export control directive to suspend access to Fable five and Mythos five to any foreign nationals Given the breadth of that order, the only choice that Anthropic had in their estimation was to take the model down entirely. Now the first phase in the absence of better information was stunned reactions in everyone casting about for who to blame. And while there might have been some partisans for either the anthropic side or the U.S government side, candidly no one was looking all that great I did an emergency episode on Saturday morning, and subsequent to that, we started to get a little bit more information, both in the form of narrative construction as well as in the form of reporting The first important post came from former AIsR David Sachs who wrote I've had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true As we know, Anthropic publicly released its Mythos class models earlier this week under the commercial name Fable fable is Mythos with guardrails. But if those guardrails fail, then you've exposed Mythos and its advanced cyber capabilities to people who shouldn't have them Keep in mind that Anthropic itself widely promoted the idea that mythos was a cyber weapon and needed to be regulated as such They ask for government regulation of mythos and champion the guardrails on Fable If there is a vulnerability, biger or small, it is Anthropics's responsibility to patch highly credible, trusted partner of both anthropic and the U SS government who is testing Fable came forward with the jailbreak of these guardrails the admin asked Jario to fix the jailbreak or redepploy the model Dario refused. In their blog post, Anthropic defended its decision by saying the jailbreak isn't serious. That is not what the trusted partner in the U.S government believe nor is that kind of minimizing language consistent with Anthropics's brand as the AI safety company It's difficult to fathom how they could claim a jailbreak allowing operability of a cyber weapon could be defined as not serious. past anthropic has always said that safety must be top priority and taken super seriously. In this case, anthropic prioritized the continued offering of the consumer model over safety In reaction, the admin issued the export control. The admin did this reluctantly It's been very surprised that Anthropic hasn't wanted to cooperate with a reasonable safety request i. e fixing the jailbreak issue Anthropics's Reaction is very much at odds with their branding and thhos as a safe AI research company The Admin's hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release The admin wants all of this to happen as soon as possible It is frankly bewildered that Anthropic hasn't wanted to comply with safety requests that it previously said were its highest priority And fininally, those trying to misdirect and tie this action to the prior DOW andthropic issues are wrong the admin values Anthropics' technical capabilities and feels that this issue while serious, should be easily resolved The Ball is in Anthropics court Everyone saw that, everyone shared it, and everyone had something to say about it There were some who took Sacks at face value. There were many more who viewed this as an act of narrative construction more than anything else And some were just outright skeptical entirely. AI entrepreneur, Eric I four. he's responded More likely story the jail brereak wasn't super serious, a situation anyone who has ever received bug reports is familiar with Andthropic thought the demand to halt the model was absurd And the federal government used the opportunity to punish and humiliate Anthropic for the prior sins of not bending the knee Anthropic has more credibility on such topics than Washington Now, if we take Sachs Post, as the modern day equivalent of a press release, there are a few things worth noting First, Sacks, in numerous cases, painanthropic is hypocritical. Not only in that they're the safety company and not taking a safety issue seriously that they had marketed Mythoss weapon as a cyber weapon and then were surprised when it needed to be regulated as such. Sacks paints the administration as being reluctant to kncap anthropic And in my estimation, paints two ways out of the situation The first is in that second to last line The admin feels that this issue while serious should be easily resolved basically saying Ball is Anthropics court In other words, anthropic resolve this and we can move on. So what's the second way out of this I don't think it should pass our notice when Sach stops talking about anthropic and starts specifically talking about Dario Amadee himself The admin asked Dario to fix the jailbreak or dedepploy the model, Dario refused. Now so far, no calls from the administration for Dario's resignation have come Although certainly that's been a subtext of some investor conversations and one strand of discourse on the internet by specifically targeting out Dario. I do think Sax is at least a little bit setting him up as a potential sacrificial lamb Now, Anthropic specific position restated from their Friday blog post, while not giving details of the specific jailbreak, they said, are basically arguing that the jailbreak that was shared with them was specific and discrete rather than universal. and universal jailbreaks are what's really important in language that was basically guaranteed to not help the situation, Anthropic went out of its way to talk about perfect jailbreak resistance not appearing to be possible today. Still, the key thing, as we try to understand what the administration saw, is anthropic is basically saying that not all jailbreaks imply the same level of risk Just because one narrow jailbreak is successful doesn't mean there's the ability to remove all guardrails Think back to what these guardbils actually were They were blocking everything from basic questions about mitochondria to any prompt with the word cancer If you manage to get the model to tell you that mitochondria is the powerhouse of the cell, that's technically a jailbreak However, it's also completely different to being able to use the model to build a bioeapon But what about this highly credible trusted partner As I was finishing recording on Saturday morning, there were the first indications of who that trusted partner was, with that information getting filled out throughout the day. Multiple outlets reported that Amazon was the unnamed trusted partner that reported the jailbreak to the U.S. government Axios reports that Amazon contacted administration officials on Thursday night to share the issue Using the precise sourced language reported by Axios, to report showed, quote, how they were able to jailbreak and access portions of Anthropics's powerful new Mythos model that pose a national security threat sources said at Anthropic that they had notified the government multiple times prior to the june ninth release, and there were no objections. However, AxiOos reports that senior admin officials fielded calls from at least five other companies across Thursday night and Friday morning. this led officials to make the decision to shut down Fabable. Contacting in trapic at one PM on Friday Anthropic sources said they received notice that they had ninety minutes to take down Fabel and Mythos due to a quote national security threat But the source also said that Anthropic received no details about the nature of the threat of that time At five thirty, Anthropic received formal notice that Fable and Mythos would be subject to export controls that barred access to all foreign nationals Around ten PM andthropic complied by taking down the two models Anthropic sources said that Dario Amade and other executives spoke with the administration after that five hundred thirty notification. According to the account provided to Axios, they laid out that the jailbreak was fairly simple, could be achieved with other models, and did not demonstrate a flaw with Fables's guuardrails Now while Axios had mentioned the administration receiving a number of other calls att least based on the best reporting that we have so far It does seem like the decision to impose export controls was based pretty much solely on the Amazon report. Despite those multiple tech companies placing calls to the administration, Amazon CEO Andy Jassy appears to have been the central figure The Wall Street Journal reports that the decision to shut Fable five down was made after conversations between Jassie and key officials, including Treasury Secretary Scott Besson Amazon for their part aren't saying much, with a spokesperson stating, as a leading cloud provider that serves a large number of private and public sector customers It's not uncommon for governments to seek our councsel on potential security risks. When they occur, we don't share details of these discussions journal included numerous details about the contents of the report source to Andrew Morris, the founder of cybersecurityirm Gry Noise Intelligence Rortedly Amazon's researchers were able to use a jailbreak to make Fable discuss security bugs in at least four software platforms Morris commented that this is information that would typically be blocked by the guardrails, but was, quote, still a long way from dangerous cybersecurity information. He noted that many other AI models can surface the same information. The difference that made Mythos a unique security risk was supposedly the ability to translate these vulnerabilities into functional exploit code. Now, Morris said that Amazon researchers had not presented evidence that they'd been able to get fable to produce this type of code, which is blocked by the guardrails There's also suggestion that the administration misinterpreted how severe the issue was writes the WSJ Jassie's calls to administration officials were viewed by some as a general warning quickly escalated into a wide commerce department ban on foreign users accessing mythos and fables Politico added further details about exactly who took part in the decision making process They wrote that Treasury Secretary Scott Besscent, White House Chief of Staff, Suszy Wiles, and White House Cyber director Seaan Karncross were the key administration figures leading the meeting where the decision was made Besscentt was apparently so critical to the AI policy decision that he joined the meeting remotely en route to an event in Houston. after the meeting, White House sources said they attempted to reach Dario Amade, but were told he was unavailable because he was attending a wellness retreat There is, as we will see, a lot of contention about that particular point White House sources said that Amadee called the White House at around one PM, but Anthropic rejected this version of events, stating it was absolutely false They noted that Amade was first requested around noon and was on the phone within an hour and fifteen minutes company made other senior leaders available in the interim Once Abmade was available, he participated in three phone calls with around half a dozen officials In addition to Besscent, Wiles and Carncross, Commerce Secretary Howard Lutnick participated in the calls together with a handful of other senior commomerce and White House staff Now at this point, it's worth noting that President Trump's name hasn't been mentioned on the list of decision makers. The only mention of Trump's involvement came from the Wall Street Journal writing President Trump later signed off on the action despite reservations about its hindering innovation As we will see, I think at this point President Trump himself is basically the only primarily innovation concerned actor in the White House anymore During the calls rightits Politico Amadee tried to clear up what he assumed was a misunderstanding He pushed back on the administration's concerns, defending the guardrails, and argued that the type of bypass that occurred, which he believed to be specific, did not pose the same risk as a broader jailbreak that would allow it to be used without any of the guardrails put in place byanthropic Karen Cross and Bessett were reportedly uncvinced by the argument One White House official said that Amazon's findings were run past the NSA, leading Karen Cross and Besscent to feel that they had quote unquote proof Government officials urged Anthropic to voluntarily withdraw the model and coordinate with the government to address vulnerabilities Amade reportedly requested more time and more information but refused to commit to taking down the model Besson at one stage warned Amadeay that he was making a bad decision. In a statement quoted to Politico, one senior White House official said, expxport controls were a last resort after begging them for hours to work with us. This was not something we wanted to do, but our hands were tied Anthropic again disputed this version of events, with one source close to the company commenting The White House gave ninety minutes to take the model down with no details on the actual threat There was never any begging or asking for them to work with us, just a declared ninety minute deadline Three White House officials commenting on the story said that Amazon wasn't the only company to raise concerns, but they didn't name the others According to political sources, in fact, it seemed like anthropics's attitude was a deciding factor One source commented The crux of the issue was the lack of seriousness that anthropic was applying to it Had anthropic taken it seriously, and rather than dismissing it as isolated move to fix or pause access, this never would have happened. 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 one point four 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 at kpmG. com slash US slash sophisticated That's kpmg dot com slash US slash 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 twelve percent 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 Seection 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 Asith the result, you go from rolling out tools to driving measurable AI value Employees move from meeting summaries to solving actual business problems, and you can prove the ROI guessing if your AI investment is working Check out section at sectionAi. com SEC Ti O n AI. com You know Assembly AI for having the most accurate streaming speech to text out there But they just went a step further and launched a full voice agent API. The idea is simple, one connection and they handle everything, the listening, the thinking, the speaking You just stream audio in and get your agent's voice response back We're talking about things like outbound sales calls that actually qualify leads. customer support that handles complex requests without a script, scheduling agents that sound like a human assistant, and you can build one in five minutes with one API And importantly, their streaming model is the best at catching all the stuff that breaks on other voice agents, things like phone numbers, emails, names, and medical terms. And for those of you who are still in experimentation mode, there are no contracts and unlimited concurrency so you can actually test it out without any friction. Head to assemblyaI. com slash brief. and try the live voice agent demo right there on the site, no signup needed This episode of the AI Daily Brief is brought to you by Out Systems, a leading agentic Systems platform built for the enterprise. Organizations all over the world are building, orchestrating, and governing agentic systemstems on the O Systems platform and with good reason OutSystems open and unified platform allows teams to architect, deliver, and scale governed agentic systems with agility. Teams of any size and technical depth can use O Systems to build, deploy, and manage AI apps and agents quickly and cost effectively without compromising reliability and security Without systemystems, you can rapidly launch ideas from concept to completion. It's the leading agendic Systems platform that is unified, agile and enterprise proven, allowing you to accelerate growth, reduce operational friction, and deliver real enterprise impact with AI Out systems, builduild your aentic future Now one thing going on here is clearly challenges around technical understanding Luda Securities Katie Mazoris wrote a blog post about the specific threat but effectively argued Anthropics point that not only was it not a big deal but that it can't be addressed without making the model worse at fixing bugs and verifying patches The softare engineer Coreory Ward wrote, Can we stop calling an LLM finding bugs in a code base that has access to a jailbreak? That's not what this is. The guardrails are not intended to prevent finding or fixing bugs. They're intended to prevent the model from being used to identify and leverage new exploits There is no safety issue demonstrated by this example. There is no evidence this is a violation of the guardrails. There is nothing to suggest this is a wholesale jailbreak of the safety measures there is thus nothing for anthropic to actually resolve with the behavior of the model It is entirely down to politics AI policy reacher Miles Brended wrote, Sounds like some folks at the White House were unaware that Fable has greater than zero cyber abilities, so thought something unsurprising was surprising Dario rightly thought this was a misunderstanding, no one in government who knew what was going on was looped in or could stop it, and people outside egged it on Part of the issue as Colin Camera puts it, lots of knowledgeable White House tech people left pointing out who Dario participated in calls with on the White House side Coolin argues the White House are way out of their depth Miles Brndig again later writes, There hass been zero indicating that domain experts at CAISI or NSA were involved. All reporting points to senior White House officials And in that context What anthropic thought was calm rational explanation clearly came off as not taking the issue seriously Remember, one of Politico's White House sources said the crux of the issue is was the lack of seriousness that anthropic was applying to it Now we are clearly getting two very different versions of events of F Friday. One from the White House and one from Anthropics. One detail that seems to be most at odds is whether Dario was actually at a wellness retreat Tech reporter Ashy Vance gave his account, writing, none of this was some weeks long back and forth. I was at Anthropics HQ on Friday reporting when all this unfolded Dario is not at a wellness retreat The Fed seem to be scrambling to try and make an example of anthropic again This is not technical, it's petty Later Ashley continued, The Feds don't like Dario Amadee because he won't do other their bidding And so now we've entered the Soviet style propaganda portion of the program, with the White House feeding every reporter it can finded with laughable claims like Dario is unreachable at a wellness retreat O the US would not be selfeating on AI since it's kind of one of the last hopes the US. has versus China. Here we are already Speaking about the wellness retreat point, Jeff Cafe writes. This seemingly minor detail included by the White House is what Scott Adams would have called a linguistic kill shot A simple, sticky idea that immediately generates a visual in anyone hearing the phrase Nobody can think about this story without picturing Dario like this. The meme that he is pointing to is Dario in a bathrobe, with two cucumbers on his eyes as someone leans over and gives him the news. Now from there it just got weirder On Saturday afternoon, we got an entirely alternate explanation that it was in fact China all along Washington Insider Publication Semaphore wrote The White House imposed export controls on Anthropics's powerful Mythos AI model partly over suspicions that a China linked group had accessed it, a person familiar with the matter said Now beyond that, the article contained almost zero new details, continuing, it's unclear how the White House learned of the issue, which organization accessed the model, and how it gained access to Mythos. But if the Chinese government had access to Mythos, it could pose national security risks to the US China could also attempt to reverse engineer and copy the model in a process known as distillation An anthropic spokesperson commented that the White House didn't raise any concerns about China during the discussions and also noted that anthropic models are already blocked in China Ashley Vance again Quick, someone say it was China, but it's really well sourced, right? Well there was a person familiar adddding more fuel to the fire that this was actually personal and about the White House's antipathy towards Pete Hagseth, despite David Zack saying that it wasn't Defense Secretary Pete Hexithh tweeted Three months ago, the Department of War kicked anthropic out of our building forever The passing day proves why that was the right move And at this point, I would argue that even if we find it abhorrent and distasteful, that policy this consequential is being made on the basis of whether people with different political opinions like each other or not The reality is that in this moment, the world that we actually live in right now, we don't have the privilege of just sitting back and finding that abhorrent. and more importantly doanthropic AxiOos published a follow up article this morning called quoting an administration official saying they screwed us. The article was titled, Personality Clashes Sent Anthropics models offline Axios quotes a source familiar with the admistation's thinking saying Anthropic has not done a great job at trying to speak to the administration and appreciate the ideological differences It's likely they just speak in different languages. Another source from around the administration argued that the admin viewed Anthropics's position at the outset as no, we're not going to do anything about it. This is not a real issue For some, the real politic is very obvious here. Bo capital Boke Rot. Much of the tech community is exposing themselves for not understanding the absolute basics of political theory. The U.S government has a monopoly on the use of force A private citizen cannot speak the way Dario speaks He will have to change his messaging or be destroyed. Category Ben Thompson wrote a long post this morning called Anthropic Safety Superpower In it, he made what I believe is a salient point about this He basically said that all of the things that people have been critical about Fable for over the last week or so for anthropics simply come down to safety The change in the data retention policy, guardrails, everything was about safety Thompson writes, Here's the thing about these safety justifications. I think they work because to anthropic, they aren't justifications The company really believes that they are the only ones who believe in superintelligence and thus are the only ones who are sufficiently concerned about the dangers that excuses decision after decision, policy after policy And confrontation after confrontation that to people on the outside Looks like a bizarre combination of cynicism and naivety Ben concludes I respect this alignment and I fear it I respect it because it is so clearly effective. The closest analogy is probably Apple which has framed every self serving action in the guise of doing right by users, and often they were So it is withanthropics What I fear, however, is that it is one thing to have people convinced they know best building a smartphone that I can take or leave It's considerably more concerning to have them building superintelligence that has the potential to rival or exceed the power of nation states or merely massive corporations The history of brilliant people convinced they know what humanity needs is a sordid one precisely because they have convinced themselves that their intentions are good justifying actions that very much are not I personally have seen this sort of thing up close and personal before. And to the extent that Ben is right, I believe that anytime people start convincing themselves that they are the only ones who are sufficiently concerned about something and the only ones who know the right remedy Things tend to get very bad very quickly And as Thompson points out O occasions of all that are radically amplified because of the power inherent in the situation RSI senior feellllow Adam Terrier argues that when push comes to shove, as interesting as it might be to debate, Whatever gotd us here, the policy is disastrous Adam writes, Because this is happening toanthropic, the temptation for many will be to say play stupid games, win stupid prizes They have relentlessly raised the regulatory temperature in Washington by inviting far reaching controls of frontier models. They made this bed and now they have to lay in it This decision by the Trump administration should not be judged on a desire for payback politics, but on the merits spepecifically, what it means for America's broader AI objectives. regard this action is truly outrageous How exactly is the government planning on even going about verifying everyone who uses this specific model to ensure compliance? that alone raises huge flags the latest executive order shifting more control to NSA, and this recent chatter about quasi nationalization and equity stakes, and now this action We are talking about a significant escalation in the politicization of AI and centralization of control over advanced computation in this country And this is all being done by an administration that had previously made acceleration in winning the great AI race a priority. We're moving backwards now. In a companion blog post Adam wrote A leading USAI company was forced to take down a product that millions were using based on non public, unexplained concerns of a few government officials. This isn't a red tape risk of the FDA, It's more like the FDA demanding, out of the blue and without explanation, that everyone stop drinking milk if milk was fifty percent of last year's stock market gains Now at the time of recording, which is right around nine AM Eastern time on Monday Everything is still in the air Last night, the Wall Street Journal reported thatanthropic had dispatched a slew of senior staff to DC trying to resolve the issue Most of the article was just a rehashing of everything we already know But we got a handful of names of whothropics entntity see. writes the journal Anthropic sent senior technical staff to Washington, including top security researchers Nicolas Carlini, Logan Graham, who leads the team that evaluates models for risks and David Or the company's head of safeguards They will meet with security experts from the government in hopes of de escalating the conflict, said people familiar with the meetings. Now it's been so little time that we're only just starting to see a public response On Sunday, a group of cybersecurity leaders, led by former Facebook Chief seecurity offfficer Alex Damos, published an open letter pushing on the administration Dear Secretary Lutnickk and National cyber Director Karancross, they wrote, We the underigned executives and technical leaders from across the United States, write to you to ask you to lift the export control directives on anthropics Fable and Mythos LLMs and commit to an open scientific and transparent process of handling AI risk assessments in the future Their argument is basically that other models are good at this stuff as well, that the multiple protections that they've built in are pretty good, that it's essential that researchers have access to those tools so they can harden their own defenses with them, and that Chinese openweight models are only months behind Now this particular letter isn't necessarily getting a ton of reach, but I will be watching to see whether there are other sorts of responses that come up around this as well Ultimately From as far as I can tell from outside The resolution to this is, like it or not, the resolution to this is not going to primarily be technical. It's going to be interpersonal Anthropic has up until now clearly thought that they could simply reason with the White House and make them see things as Anthropic does Clearly not the case And whether they wanted to be in this position or not, Anthropic can no longer play the scrappy startup They are one of two leaders in the most consequential industry, certainly for the economy, if not the larger world as well Gott to play ball with the government they have. not just the government they'd like. Responding to news about anthropic senior technical staff going to Washington, investor Melinda Chu wrote, If Daro Amed iss not on the plane, nothing will change Now it is highly likely by the time you are watching this, it'll be out of date, but of course, we'll be back with another update very soon For that's going to do it for today'say Idaily brief. Appreciate you listening or watching as always, and until next time Pace.