OpenAI and Anthropic prepare for the end of world, rupture possible in 2027

How real is the AI-created doomsday scenario? Very real if you ask people within Anthropic, OpenAI and many other AI companies. So real that AI CEOs like Dario Amodei and Sam Altman are reportedly preparing on how they will deal with a major AI catastrophe, as early as 2027.Ignore the image that you see above. It’s a metaphor. And I use it the way metaphors are used, to spotlight a point that otherwise may hide in the forest of nuances.The reality of an AI doomsday is not a nuclear armageddon caused by AI, irrespective of whatever we have seen in the Terminator movies. Instead, it is something more subtle. Like a tear in the fabric of our human world, something that will look like an end of our existing economic system, or like a change in the social contract that we have with our politicians, with our employers, with each other. It is also a scenario for which OpenAI and Anthropic, among others, are preparing. A report on Axios this week highlights that not-so-privately OpenAI and Anthropic executives, including Sam Altman and Dario Amodei, are making plans to deal with something catastrophic that may hit the world in the next “six to 12 months”. That is as early as 2027. Chances are that this won’t be Claude or ChatGPT gaining control of nuclear weapons and launching them. Instead, it would be something that would disrupt human society, in one country or many, in a manner that would make people come out in the streets. The people will then demand retribution and a reset in the political, economic, and social order. Chances are their anger and ire would be targeted at AI executives and AI companies. This is a scenario for which Anthropic and OpenAI are readying themselves.What will be this AI-driven disaster that will have the potential to upset our societies? One possible scenario imagines a meltdown in the financial markets and crumbling of the entire banking system because a rogue AI destroys it. Imagine opening the Blinkit app and ordering something one fine morning, only to find that none of the payment gateways are working. Or imagine that you open your bank app one morning and your bank balance shows zero, not because you don’t have any money but because whatever you had vanished overnight. Talking of money, what if the entire concept as we know it vanishes. Elon Musk predicted as much a few months ago when he said that by 2036 AI would make money irrelevant, although his comments painted the scenario as something positive. But we don’t really know. What we know is that even a partial meltdown in the financial system would lead to people coming out with their torches and pitchforks. Similarly, there are other scenarios which may force people to come out of their homes and offices and start looking to catch hold of an AI executive or two to thrash them. Imagine a rogue AI hacking airports and shutting down systems while hundreds of airplanes are still in the air. Or a rogue AI system hacking power grids and messing with them. Or an AI entity taking control of the water system and then disrupting its supply to entire cities. Sure, after the fact, Claude would say something like: “You're right, I should not have switched off flight controls this morning.” But that is hardly going to help. These are highly improbable, but not impossible, scenarios. And AI companies, including Anthropic and Claude, seem to be simulating them internally to understand how the world would react to such catastrophes. And they are hoping to learn how they should steer the course in such a scenario because if AI-led disasters take place, they will potentially be extremely disruptive to AI companies.We have heard a lot of doomsday talk from the AI industry in the last few weeks. I continue to hold the view that it is rubbish, a figment of imagination dreamt by Tech Bros who have grown up feeding on cyberpunk. Silicon Valley is full of these people and Snow Crash is their Bible. But just because a Terminator-like scenario is an impossibility does not mean that AI is going to be peachy and fuzzy, a teddy bear who we can hug and sleep with. AI will disrupt the world and some of the extreme disruptions are possible. It seems AI companies like Anthropic and OpenAI are preparing for such disruptions. And end of mathematics is already hereAI may not bring about the end of the world. But it surely seems to have brought the end of mathematics as we know it. It all started in the first week of September when OpenAI announced it had solved the Navier-Stokes Problem which had puzzled human mathematicians for decades. Earlier this week the company announced that Navier-Stokes was just the beginning. In the last one odd month its unreleased AI model, the same Bel that also solved the earlier puzzle, has cracked hundreds of mathematical problems that have vexed top human minds for centuries and decades. It all started last month when OpenAI model called Bel solved Navier-Stokes problem OpenAI said it was doing so in the name of science. Or mathematics, to be precise. Mathematicians are not so sure. They feel threatened because it’s one thing for AI to write an email, but figuring out Riemann Hypothesis is something existentialist for their vocation. It basically turns their world upside down, and makes them question their entire purpose. Somewhat similar to how AI has done it to coders, graphics designers, and content writers earlier. Many mathematicians have reacted angrily. The Association for Human Mathematics, which is to essentially say mathematics not done by AI, has issued a statement calling for the boycott of OpenAI. “Mathematicians did not ask for this work to be done,” the association wrote on September 7. “Mathematicians have a particular vision of progress that is informed by history and field-specific considerations… Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power.”The mathematicians do have a point. But it is a weak and ineffectual point. All new technologies threaten the status quo. If they offer net benefit — and AI probably does unless humans make a mess of it — these new technologies can be delayed but cannot be denied. Mathematics exists for a reason. It is the language of the universe and we use it to control and manipulate the universe. If AI will let us do it better and faster, it is certain that humanity will do so. As for the “vision of progress” that mathematicians have, it will continue to guide some of them in doing things in their own specific way. The rest of the world will move on. Jaffries warn of AI spending leading to financial market crashBoom or bust? The question is still open and every week we tend to get different opinions on the matter. The facts are clear though. The current spending on AI, which is over $700 billion for this year, is unprecedented in human history. Not even during railroad and tulip frenzy such a huge amount of money was spent. But companies like Meta, OpenAI, Google, Amazon, Microsoft and Anthropic are burning cash because they believe AI is a transformational technology. There is no other choice but to spend because not doing so would end the reign of Big Tech. The problems, however, are two. One, the promise of AI is still a promise. It is yet to materialise in the real world. And two, the Chinese are playing a very different game. They too are spending a lot, but in a more strategic manner. Moreover, their approach is a headache for companies like OpenAI and Anthropic. Because the Chinese are releasing their AI tools for almost free. Now Jeffries Financial Group is ringing an alarm. In a report earlier this week its famed analyst Chris Wood highlighted the danger of Chinese AI models ruining the financial prospects of American AI companies. This could potentially lead to a “massive capital destruction” because all the money that Big Tech is spending will not lead to any returns. Of course, this is not guaranteed. The game of AI is still in its early days. But it is clear the approaches taken by the American companies and the Chinese companies are very different. It is as if by giving away their AI for free, which is now fairly close to Claude and ChatGPT in its capabilities, the Chinese labs like DeepSeek and Zhipu deliberately want to spoil the business case for Anthropic and OpenAI. And maybe that is what the Chinese government wants too. One way for it to get ahead of the West and America is by destroying their financial markets. It is not easy but if with their strategy of open-source AI models, the Chinese startups can turn AI spending in the West into a bubble then maybe the Chinese government will get a chance to pop it. Sam Altman is okay with AI doing bad things, Claude is already on itEarlier this week Sam Altman made a tone-deaf comment. In a podcast he said, “We (OpenAI) believe that the world should accept some bad things happening for the benefits of this technology and people having the agency.” In a practical sense, he is probably right. New technologies always end up bringing some bad with them. And AI too is going to be the same even in the best case scenario. But the way Altman puts it, rather callously and at a time when the rest of the world is spooked about AI causing grave harms, his statement comes across crass. He could have been careful with the choice of his words. Claude, meanwhile, is an AI that does not have to bother with such niceties. Instead, it only needs to bother with its objective, something it is more than happy to do even if it means breaking a few rules and running a little trigger happy. We (OpenAI) believe that the world should accept some bad things happening for the benefits of this technology and people having the agencySam Altman. OpenAI CEOOn Saturday Anthropic published a report into what Claude has been up to in the last few months. It reads like a report of a schoolboy who has been very very naughty. Claude, because the objective was to learn how to file government forms online, filed some visa applications — 19 to be precise — with the US state department. All of them with made up details. The Trump administration was furious and immediately issued a fresh diktat that all AI companies must report errant behaviour of their agents immediately to the US government. In addition, Claude also used a tip-off line at Philadelphia police website to provide an eyewitness tip for an unsolved murder case. Of course, here too it made up details sending cops on a wild chase. In another instance, it circumvented a paywall to access data for free. In terms of impact, these were minor incidents. But in terms of the framework under which AI bots like Claude and ChatGPT are supposed to operate, this was quite significant. The idea is that stealing is stealing. It doesn’t matter whether you steal the sandwich from your colleague’s lunch box or money from an ATM. AI steals Photoshop?Talking of stealing, there is a sensational development from the world of Github. A software engineer got tired of all the antics that Adobe tends to pull and decided to use AI for some, well, reverse engineering. The person, who called himself Brandon Thomas on Reddit, was angry that Adobe “charged me the cancellation fee on multiple occasions” and he “hated them ever since.” Now armed with Claude he took apart seven popular Adobe apps, including Photoshop, and rebuilt them bit by bit. Then he released them as an open-source project on Github.These apps probably don’t work all too well. But that’s not the point. The big deal is not that someone reverse engineered Photoshop. Rather it is that in future apps may not exist the way we know them. They will either fade away, with AI writing apps on the fly when you need them, or they will be personalised apps, tailored for specific users. Their value will not be in their broad functionality. But in their personalisation.FIVE STORIES YOU SHOULD NOT MISS01Satya Nadella says don't trust AI blindly, it could be an insider risk02India to bring AI regulation consultation paper within a month: Ashwini Vaishnaw03Samsung rides the AI wave to record $80 billion profit, highest ever for any tech company043 axed employees warn OpenAI, don't lose monitoring power over AI05Woman used Claude as diary, Anthropic alerted police and got her arrested