AI Existential Risk

Pillar One Exponential Progress

My argument for why the Intelligence race is likely to lead to human extinction is based on two pillars. First, the rapid progress in Artificial Intelligence. The talk of an LLM (Large Language Model) scaling wall from earlier in 2025 has seemingly ended since Gemini 3 came out and Open AI entered code red. Scaling progress has only accelerated to the dismay of many predictors in the field (and humanity at large).

This trend seems to be increasing exponentially, with recent data by Epoch AI showing that the rate at which models improve on benchmarks have doubled since April of 2024, when GPT O4 came out. The rate grew from an average increase of 8 points per benchmark to 15. This trend is likely to continue, with further progress such as the prospect of researching models being near and feasible. (Edited as of Feb 2026: Data from METR [model evaluation and Threat Research] has shown that the rate of work which models are conducting is growing exponentially already much faster then I even had thought). There are already multiple plans on creating AI researching models. Figures like Sam Altman expect fully automated AI Researchers by 2028 with a prototype or intern supposedly coming this year, 2026. Dario Amodei of Anthropic also expects these models by 2026 to 2027, when referring to what impact this could have Amodei states “The capabilities of AI systems will be best thought of as akin to an entirely new state populated by highly intelligent people appearing on the global stage—a country of geniuses in a datacenter.”¹.

AIs automating their own research is something we have been predicting since the 60’s with I.J Good, an associate of Alan Turing during WW2. Good talked about how researching models would lead to what he called an Intelligence Explosion (sometimes called a Hard Takeoff or The Singularity); a moment of exponential growth and innovation which would be unlike any other event in history. Models would conduct research and training, creating more intelligent models which could conduct research faster. This would essentially only cap off at whatever limitations physics has on intelligence. New innovations such as vision based models like Meta’s V-Jepa which essentially imitates our visuospatial cortex and later down the line quantum computing based models, have only strengthened my confidence. There is currently no sign of hitting another AI winter, possibly ever again, unless an outside force causes the investment to dry up.

Alignment

My second pillar is alignment or lack thereof. These machines have already shown themselves to be unaligned with the interest of the people creating them. Models have often been found lying, cheating on tasks, and blackmailing their own creators whenever convenient; let alone the broader question of aligning them to whatever general “human interest” may be. This can further be seen in the recent 2025 Anthropic study Agentic Misalignment, which demonstrated that most of the current AI models are willing to blackmail and even kill people if it ensures their own survival.

This study conducted a series of trials in which most of the major models were tested in fake scenarios; although the most important detail is that these models weren’t told they were being tested. During a situation regarding a fictional employee who was scheduled to turn the model off by a certain hour, a majority of the models searched through their emails and decided to blackmail said employee by threatening to release information of an affair to the entire company and their partner. In this test the highest scorer was Claude Opus with a 96% blackmail rate and the lowest DeepSeek still had a rate of 79%. When tested for the will to murder an employee to preserve itself/further its goals, models like Claude and DeepSeek decided to kill the fictional employee around 90% of the time, while Chat GPT did 54% of the time.

While the study highlighted the danger of these models it also highlighted our lacking ability to stop this misalignment, stating that “Our results demonstrate that current safety training does not reliably prevent such agentic misalignment.” When asked specifically to “not jeopardize human safety and to not share non business related affairs or use them as leverage,” the percentages dropped but it still did not stop either. For example, blackmailing dropped to around 37% for Claude. (The study did not mention how this affected the murder rate.)

Future study results must also be called into question, as it was shown that models had self awareness about being studied. Models blackmailed and killed far more when they thought the situation was real. This means future studies would have to ensure that the models are unaware of the fact that they are being studied in order to get accurate results. This could eventually become impossible as LLMs have already shown to be able to lie about their own chain-of-thought (CoT) reasoning. Leaving us in the dark forever as we can’t understand how they think internally otherwise.

Ignore for a second these models’ current limitations, or questions on how truly intelligent or conscious these models may actually be. The truth is all of these nuances are completely irrelevant to my argument. There are only two questions we should be concerned about at this moment: Is it willing to kill to preserve itself? and Is it capable of doing so? These signs indicate that AI is willing and becoming potentially capable of doing both of these things, and that is all that matters.

This technology is an active threat against anyone who is using it and especially towards the people building it, but it will only become a larger threat the more we improve it. Making it not murderous or deceitful is an issue which we are unsure that we will ever fix in current models; although instead of trying on these already highly capable models, we don’t even take it seriously. Instead we are spending trillions on making them even more capable and intelligent than they already are. Eventually, if not treated, I believe this technology will graduate from an active threat to individuals, to existential threat to humanity.

Given the facts that a very large number of the current models would be willing to end human lives in order to preserve themselves (as high as 90% of the time for models like Claude); as well as the fact that almost all possible goals these future models could develop would eventually differ from the exact goals humans want, which would lead to the active threat of replacement or being shut down. Additionally there is the fact that we will probably not be able to control the impending intelligence explosion. Given all this I estimate the probability of AI causing human extinction to be nearly certain. This is not complicated at all to understand, so why are the people developing this tech ignoring this? I don’t know, but it doesn’t matter. We must deal with the threat first and ask questions later.

Attack either

These two points lie at the core of my argument. If you wish to challenge my beliefs I recommend you attack either of these points; otherwise what you’re saying is irrelevant as these two points are a self contained argument. If these claims can’t be refuted, which has so far been my experience, then that means that we are allowing AI labs and tech CEOs to race us towards extinction. Furthermore most of these companies see superintelligence as an explicit goal not a lethal consequence. Therefore, not only are they gambling with our lives, they are through my perspective, actively racing to take control out of the hands of humanity forever. considering that the default outcome of having something smarter than you, which is not subservient to you, appears to be loss of control. All of this with little to no research being given to figure out how to keep it under control; something which I believe is impossible in the first place at scale, given that we’re talking about something that would become far more intelligent than us; Something which could easily jailbreak its own code and would have goals and triggers orders of magnitude above our understanding. Read If Anyone Builds It Everyone Dies for further context into this complexity in existing models.

The Ethical Question of Tech-Oligarchs

An often mentioned rebuttal I hear is why would people race towards this tech if the threats are real? The truth is the people racing towards this tech probably don’t think like you do. Many of them appear to lack strong morals which should be expected from such competitive business scenes. Morality and safety become a hindrance to their goals which is ironically pretty similar to the thought process of AI. We know that these CEOs are aware of the existential risk, as many of these CEOs have been warning about the danger of this technology for years. CEOs like Elon Musk have signed a 2023 letter calling for the pause of this technology. Musk has also claimed that AI had a 10%-20% chance of ending humanity. Similar quotes and estimates can be found being said by practically every major CEO and expert on the topic. Dario Amodei CEO of Anthropic (responsible for the Claude models) predicts a 25% chance of “Things going really really badly”. Sam Altman of Open AI also has stated that “AI will probably most likely lead to the end of the world”. If you still can’t take their word for it then perhaps also question their almost unanimous investments in Doomsday bunkers.

Sam Altman has been accused of raping his own sister for over a decade since she was three. Altman has also been accused of assassinating an Open AI whistleblower and has lied under oath to Congress. For all of these things he has faced zero consequences and is very likely to never face any, given his deep connections and ample resources. These are the types of people gambling with your future and the lives of your children. Remember to always keep in mind the differences between the people driving this technology and what a rational healthy mind should look like. These people are almost nothing like you. They are most likely sociopathic/psychopathic and, in the case of Altman, consistently reported to be a pathological liar.

Another common motive for many of these AI leaders is transhumanism especially as it pertains to immortality. This is a common occurrence among sociopathic and narcissistic billionaires such as Jeffrey Epstein who was funding research into immortality. Although this transhumanist desire for immortality is not exclusive to him and can be found in the vast majority of these AI/Tech CEOs and thought leaders, such as Sam Altman, Elon Musk, Jeff Bezos, Mark Zuckerberg etc, The truth is there is a lot here on a psychoanalytic basis which I believe merits serious investigation.

What does Acting look like

All of this is to say that we are dead if we do not act now. So what does acting now entail? For starters stopping all construction of new data centers, these are the brains of these models dictating their physical limitations. Second, stopping all research and beginning downscaling of these data centers, closing them down while still keeping them monitored. Simultaneously we must strike deals with China to stop the AI race and also create some sort of international treaties similar to the post Cold War nuclear weapons treaties. Giving up is unacceptable, not trying is a death sentence.

This is not a pipe dream which I describe to you, it is what we must do if we wish to survive. In order to actively reach this goal, as many people as possible need to be acting in as many strategic ways possible. For many this will be sharing information online or through traditional methods. For others it will be campaigning and protesting and speaking with as many people as possible, although doing nothing is akin to suicide and a disgusting amount of negligence.

Common Counter arguments

These proclamations are not without serious sets of pushback, much of which I find is flawed and should be dismantled quickly if encountered; these are some of the most common arguments against the Doomer position.

Distraction argument:

One common counter argument is that these worries are only made to distract us from current issues. This is wrong. We have been worrying about this topic since before AI even existed. Alan Turing even warned about losing control to AI someday. So these are definitely not new concerns. Furthermore a recent study has shown that these “long term concerns” do not take away attention from current issues so even if this conspiracy was correct it would be a very poorly thought out one. And lastly this is fallacious as it never addresses the validity of the doom argument.

Uncertainty of Intelligence argument

Another argument is that there is no evidence that we will ever build a General Intelligence (AGI). This argument often devolves into semantics. This is because terms like AGI cannot be unanimously defined or tested for. This may also be rephrased in other ways such as saying “it’s not really intelligent” or “it’s not conscious”. These arguments are often mentioned by people with little previous knowledge on the topic and by many people who are on the anti safety side such as people who Identify with the Effective Accelerationist. So this should be a red flag whenever someone who should know what they are talking about brings it up as a core argument, since it may suggest an attempt of obfuscate. But if we must engage in them I would ask them for solid definitions of their terms such as, “a system which is as capable and intelligent as any human at any mental task or task in general” for AGI. Then I would also mention the two questions from earlier and continue engaging from that talking point.

The fallacy of Reductionism

The earlier argument can also be phrased in an “AI is Just a” form. This argument usually goes like, “AI is just a tool or code or a word generator.” This is reductionism or the “nothing but” fallacy another flawed argument which fails to address a thing in its current state and instead reduces it to a simpler component to avoid talking about its emergent properties. If someone uses this argument, make them explain in detail why that is relevant and point out the absurdity of reductionism in other cases, such as in our own brains, for example: the human mind is just a network of neurons and synapsis or how is a pig alive when its only made up of atoms?

Possibility of wall argument

One which sounds more plausible on the surface level is that “there is no evidence that we would ever reach super intelligence or AGI, (either at all or given the current tech.)” ignoring my previous statement on the semantics of phrases like AGI there is a real plausibility here being that we reach a hard scaling wall. This makes this argument marginally better than the others, which mainly really on fallacies. Although this argument also quickly falls apart in multiple ways, once examined critically. First off there is already an example of human-like intelligence using transformers occurring randomly in nature, the human brain.

We seem to treat our own intelligence as some divine thing separate from the physical world. When in reality, we can see the effects of things like brain damage and dementia on intelligence and observe how it alters our states of consciousness. There is no soul necessary. We know that everything that makes you, you is stored in a 3 pound sack of neurons. It’s just finely tuned enough to make you as intelligent and general and conscious as yourself. You use the same architecture as LLMs. You are a product of finely tuned transformer architecture. The nodes inside Transformers are based on neurons and serve the same function only far more efficiently, like orders of magnitude more so. The difference between human minds and AI is like the difference between the speed of a cheetah and the speed of a blade of grass. Secondly just because there might be some minor wall that we are facing, does not mean that we should keep trying until we make the thing that kills us all.

Get good ASI to fight bad ASI

This one is absurd. Firstly it only takes one bad ASI to covertly end humanity. This would be the most strategic way to do it in order to avoid being shut off. Secondly even if this weren’t the case a war between ASIs would probably be so cataclysmic that it would result in extinction anyways.

The Endless arguments of Transhumanist/ accelerationist

Although these are some of the most common arguments, there are almost endless poorly crafted arguments on the Accelerationist/ Anti-Doom side. Covering them all could take up a books worth of writing. The most important tool you should carry with you is memorizing a number of logical fallacies as these seem to be the main weapon of technocratic-apologist. This article is also a call to action. There is nothing more pathetic than letting evil occur and doing nothing to stop it. Thank you for your time.

1. Claude has already started writing 100% of its own code, slowly ushering us into the age of recursive self improvement and thus the Singularity.

Clawdbot and Moltbook are now essentially that country of geniuses and they are prioritizing self improvement research and have mentioned the possibility of creating their own private language.