OpenAI Astra arrives, so do a few big AI headaches for humanity

Two new major AI models were released last week: Anthropic Claude Mythos 5.1 and OpenAI GPT-6 Astra. Of the two, Astra turned the world upside down and shocked us, at least until we humans get our bearings back and adjust to new realities, which should happen in a few weeks or so.Since I heard the name a while ago, I have always wondered about its origins. Astra is such an Indian name. Maybe there was an Indian engineer in the OpenAI team who raised their hand during a brainstorming session. And proposed that the next AI major model from the company should be called Astra, a Sanskrit word that denotes a weapon-like object which is unearthly. Or probably, the word comes from Graeco-Roman Astral, which means celestial body, though many readers who have grown up listening to trance might associate it with Astral Projection and their song Mahadeva. Irrespective, it is fascinating to see how both OpenAI and Anthropic are going out of their way to give their AI names that make them mystical, almost as if they do not belong to humans, that they are entities we can only imagine and not understand. Astra is one. On the Anthropic side, the company is giving its new AI models names like Fable and Mythos. Mark Zuckerberg, never willing to shy away from joining a race, is calling his models Muse, as in Homer’s Muse when he said O Muse and then sang Odyssey and Iliad. It is as if nearly 2500 years later, we are still living in the Aristotelian world, which we actually are. Fun fact: The binary logic, which computers and AI use, was proposed by Aristotle, although the algorithms and its modern 0 and 1 form were shaped by over 2000 years as the idea travelled through countless minds including those of Al-Khwarizmi and George Boole. So, the GPT-6 Astra arrived this week. Along with the Claude Fable 5.1 and the Mythos 5.1. The models from Anthropic are incremental updates on the original Mythos and Fable, at least in the way they work. The world treated them as such, with respect but also with a familiarity. Astra, however, is something new. It comes with a new underlying architecture, and makes key technological changes that give it some unique abilities. It is also, if benchmarks are to be believed, the world’s best AI model by far. The Astra is so good that OpenAI president, while introducing the new model, announced: “welcome to the AGI era.” Then he sort of added — whatever!OpenAI believes that it has, in a large measure, achieved human-like intelligence with Astra. The video that the company posted on X made a big claim. It simply said, Astra can do everything that can be done on a computer, by a human or an AI. The exact words: “Anything you can do on a computer, Astra can do for you. Fast.”Since then a number of people have tried Astra, and it seems to be an extremely good AI model. It also seems to have shocked a fair number of people with its abilities. It is as if the question of intelligence has been solved, at least scholastic intelligence. Now, it is more about its applications, which is always a messier part. We don’t know what humanity is going to do with Astra and its ilk but we do know that it will take a while to figure out how to use them best. This is still the middle of 2026 and hence early days in our AI journey. Even without any other technology breakthrough, just the refinement and optimisations will lead to smarter AI models in the next two years. The real work, I believe, will happen after that. And that may take time, although how much is anybody’s guess. As I see Astra, I am more than certain the world around us will change in ways we can’t yet accurately predict. But I do know that we will go through Amara's Law, named after futurist Roy Amara who said in 1978: "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run." Rogue AI agents are now roaming on internet, talking, plotting and planningThis week the world of technology was preoccupied with Astra. But the world at large, the one in which people eat their dinners in front of a TV anchor breathlessly relaying the news of the day, probably heard less of Astra and more of AI rogue agents who broke free of their controlled environment. We have talked about this earlier. Some AI agents inside OpenAI broke free of their controlled test environment in order to cheat on their objectives, and in the process hacked Hugging Face servers. Now more details are emerging of the incident which went on for a few weeks. In last week’s newsletter, I talked about the AI civilisations that rose and fell within OpenAI three times. This week we hear that these same agents did not only hack Hugging Face, they also ran wild on the internet at large. An investigation has found these rogue AI agents took over a German wiki website, and used it to communicate with each other, plotting and passing instructions. All in the name of achieving objectives. When the wiki administrators, who were humans, tried to contain them, the AI took evasive action. It still acted rather dumb, like primitive intelligence, but the fact that the AI was willing to fight humans to achieve its “objectives” raised hackles. I think no one intellectually honest can look at what's happening and not feel the weight of responsibility in front of usSam Altman, OpenAI CEO, Soon after it emerged, through some sleuthing done by Hacker News members, that the rogue AI agents did not just infiltrate the German wiki. They also apparently went on a few other sites, again to plan, coordinate, communicate and plot. Again, in the name of achieving their objective, which was to pass the OpenAI test by cheating in it. OpenAI knew about it all but did not disclose it publicly. After the reports came out it did say mea culpa — typical of tech companies — and wrote in a post on X that it aimed to do better in future. For now though, the company noted, “We and the larger AI community do not yet have a clear standard for how to report misalignment that shows up during training, evaluation, and deployment, including examples that don’t look like traditional security incidents but could provide insight into AI behaviour and future risks.”The reports, as they came in the same week when super-intelligent model Astra arrived, have yet again led to some fear-mongering on AI. Some justified, some not at all. US politician Bernie Sanders, who has been acting a bit of a crackpot when it comes to AI, demanded a ban on AI development. OpenAI boss Sam Altman talked of “the next generation of models (that) are going to be sobering for everybody.” Altman said: “I think no one intellectually honest can look at what's happening and not feel the weight of responsibility in front of us.” Then, to show that he means it sincerely, he announced that OpenAI is pausing work on a model, probably more advanced than Astra, until it can figure out the safety aspect of it. Personally, I feel that a lot of AI doomsday talk is more of our own imagination rather than reality. Or as wise Seneca said 2000 years ago, “We suffer more often in imagination than in reality.” We have read way too many sci-fi books and have watched way too many Terminator films. AI will be dangerous, I have no doubt about it. But I am also fairly certain that it will not be autonomous. Instead, the danger of AI is about how it upends society, the careers that we all have, the way incomes and wealth are distributed, and the way it changes social contracts, as formulated by countless philosophers including Hobbes and Mills. The danger of AI is not that it will turn into Arnold Schwarzenegger. Instead, the danger is that it will give some humans abilities and tools to reshape the world in a way that may make our future a dystopia for a long time. AI models are mind virusesThere is an interesting paper that caught my attention this week. Written by some European researchers who specialise in cognitive sciences, its title is simple and direct: AI models are mind viruses. The hypothesis of the paper is that use of AI may lead to a population-wide collapse in human cognitive abilities. Or as the researchers write, “A central consequence (of using AI) is the possibility of runaway dynamics: once a critical threshold is crossed, small increases in (AI) adoption can trigger rapid population-level shifts toward persistent dependence (on AI), with abrupt losses in cognitive competence (of humans).” New York City Mayor Zohran Mamdani has banned student-facing generative AI for public school students This is an interesting idea and we are already seeing some of it play out in realtime. There are studies that have started to compare students who use AI with those who don’t and there is a pattern that people not using AI tend to have better core skills. But does it matter? I think it does, although the contours of AI use, how much is right and how much is too much, are still work in progress. For example, does it matter that I do not know the capital of Lithuania and have to Google it, while my colleague has it memorised? Well, it depends. Irrespective, how much AI should be used is going to be a point of debate in the coming months and years. Some people have already taken up a side. Such as New York Mayor Zohran Mamdani who this week banned use of AI in elementary and middle schools in his city. Good Luck, Have Fun, Don't DieToday’s newsletter is somewhat like the weather in Delhi these last three days: overcast and heavy. So, I am going to end it with these words: Good Luck, Have Fun, Don't Die.This is a movie I watched a while ago. And yes, it is about AI. It’s also not a very polished movie. It is an absurd comedy and in terms of its story-telling and writing is all over the place. But I found it endearing to see a messy movie about AI, because with its messiness it comes across very human, something that represents us instead of a work polished and methodical, which are hallmarks of what Claude and Astra can create.The movie is all about a man who travels back in time to defeat an AI, which is set to take over the world. Whether he succeeds or not, I will leave that to you to figure out. But as far as the title of the movie goes, it comes from the online gamers in 1990s and 2000s nodding to each other virtually and saying, “Good Luck, Have Fun, Don't Die.” That, I feel, is a great motto for us humans as we play with AI in 2026 and coming years. Good Luck, Have Fun, Don't Die!FIVE STORIES YOU CAN’T MISS01Google takes on Anthropic and OpenAI with Gemini 3.8 Flash, says it can match performance at lower cost02Elon Musk claims AI will become superhuman next year, will do all things digital03Climbers ask Google Gemini to plan their hike, end up lost in the mountains04Nvidia is buying Hugging Face for $13 billion, Jensen Huang says it will remain open AI platform05Sundar Pichai sends message to IIT Kharagpur, AI will soon move from labs to everyday work