Sep 16, 2026

Are we speedrunning an Oppenheimer sequel?

In recent days there has been quite some stir in the AI community: AI is going to destroy humanity, we need to stop.

---

It all started with the following post:

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

You can check the full thread on X, but if you don’t have an account (Elon needs more signups), you can use nitter.

This created a lot of debate and the tweet went viral, he didn’t have any posts before this one and got 301k followers and 171M views on that post. He even went to an interview to talk about the dangers of AI, why he resigned, what other coworkers should do and more. But he continues to hold OpenAI stock, so if you really got out of the AI race why keep OpenAI equity? After all these warnings on how dangerous it is, why do they continue with it? Maybe it’s just marketing and more marketing. Some people, myself included, think this drama is surprisingly convenient for Anthropic. It’s happening right as OpenAI dropped a model claiming to solve Navier-Stokes, a problem unsolved for years. All this doom discussion has overshadowed OpenAI’s big moment.

Is it really dangerous?

We don’t really know, nobody knows and it seems like they don’t even know because according to a blog post by OpenAI in 2019, GPT-2 was too dangerous to be fully released to the public so they only allowed its users a reduced version of it, after nine months they released the super dangerous full model and now we even have an open weight model gpt-oss released by them in 2025 that anyone can run on their computers. Anthropic developed a model called Mythos (nice name after what happened with it) that was supposed to be super dangerous because it could find important vulnerabilities in any software and was a threat, but then again they ended up releasing it with the name Fable 5 and “super secure guardrails” (the model probably pinky promised them he will behave good) that not even their red team after 1,000 hours of trying to jailbreak it succeeded. Is Fable 5 now secure and cannot be jailbroken? Obviously not, after 48h it was jailbroken and the U.S government decided to shut it down, but guess what? it got released anyway after some time because it was not a threat to humanity. I think current AI is not dangerous. It does not take actions on its own like if it’s some kind of terminator, LLMs just do something depending on their input. It’s all pure marketing but the threat is real if we really achieve AGI at some point. However, why would the AGI not kill us? AI will be superior to us, it learned from us and what do we do to inferior species? We make with them whatever we want like for example with pigs, we eat them, because we are better than them and they cannot do anything to avoid it. Why would AI treat other species as equals when we don’t? Maybe it lets us live since a machine doesn’t have biological instincts to survive or maybe not. Nobody knows.

Some people want all humanity to die

Richard Sutton, father of reinforcement learning, argues that we should embrace AI succession and that it is arrogant for humans to think we should remain the dominant species forever. We also have Nick Land, father of accelerationism movement, who believes that technological intelligence is a force of nature using humanity as a bridge and that being replaced by AI is just upgrading the hardware.

Threaten to kill a person: you are crazy.

Threaten a million people: you are a dictator.

Threaten an entire species: you are a visionary.

Author: Some post on the internet that I could not find again.

Let’s regulate AI so we don’t have any competitors

After all this debate this week Dario Amodei made a post about slowing down AI and asked for regulation, saying that Anthropic will start giving the same access to external evaluators as their employees, so they can supervise their AI models. He says that is similar to what is done in banking but AI is not the same as a bank, there is currently a race to achieve AGI, every AI lab wants to be the first ones to reach it and the reward for it is very high, so how can we trust that these external evaluators are going to really have the same access after all the lies they constantly tell?

Embedded Evaluators. Each frontier AI company commits to giving ongoing, employee-like access to a team of embedded third-party evaluators (such as METR), whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes. This is the key step for verifiability of any pacing commitments, and has precedent in the banking industry, which sometimes involves regulatory “supervisors” embedded along with employees. Anthropic is unilaterally committing to this step now. We intend this to be part of a broader push to redouble efforts on our safety and alignment work.

Sam Altman replied saying OpenAI will start doing the same, and Elon Musk too. The post also talks about international cooperation and that all countries should coordinate safety guidelines and find a way to verify their compliance, something that in my opinion is going to be very difficult. We are in a race like the space one, so do you think a country like the U.S would trust China with slowing down? Or that China will trust the U.S? No, they will not. Companies will continue doing the same, but instead of disclosing it to the public they will keep it for them in case the other companies or countries are cheating too.

Let’s postpone the IPO for next year for the safety of everyone

Sam Altman in an interview this week said:

Given everything happening with safety, right now would be an ill-advised moment to go public.

So now that people are concerned about safety, you suddenly agree with safety concerns and want to postpone the IPO? Or maybe you cannot do an IPO right now? Because there is probably something they want to hide. In April 2026, CFO Sarah Friar opposed an IPO because they weren’t ready for the financial requirements of a public company, while Sam was trying to have it as soon as possible. They probably realized they are not ready and used all this discussion to say: look how good we are that we are postponing it due to safety concerns, but look at Anthropic, how bad is it that they are not postponing it. And while Anthropic is supposedly concerned about security and asks for regulation, they have not postponed the IPO and are preparing everything to go public on the NASDAQ. Why would a company ask for regulations when they are racing for an IPO? Regulations are never good for business: they usually make companies spend money on lawyers and restrict what they can do to earn money. They are usually to protect the consumers, then why agree to them? Because they are the big ones, they can afford to pay lawyers and to comply with anything, making small AI labs struggle with regulations and reducing competition. They are afraid that someone will surpass them and the IPO does not go well. I also think that maybe they are scared of China AI models, so this is their last chance to stay on top. If we check the most used models on OpenRouter, Chinese models are very high in the ranking. On top 3 models on code generation we have the following:

1. Claude Opus 5
2. GLM 5.3
3. Kimi K3

The first one is from Anthropic, the other 2 are from Chinese companies that publish open weights models so that anyone with a powerful enough computer can run them, and this is a threat to their business model because these models have a very low price. In the following table we can see their prices per million tokens.

We can see a huge difference in price and the capabilities of closed source models and open source ones are closing their gap every day with a way lower price.

Conclusion

In my opinion, current AI models are not a threat, the threat is the people who use them, just like a knife is simply a tool you can use to cut bread or to kill someone. AI CEOs are just asking for regulations so they face less competition and hyping up how dangerous their models are so you think they threaten humanity like a Terminator, pushing you to buy their models and not the Chinese ones. The danger is real, but they are only using it for profit. Please watch the movie Oppenheimer: he realized his mistake once the bomb was already built. It was an awesome movie, but we don’t need a sequel on how AI labs regret building AGI and how they didn’t know how bad it was. Be like Jacob, and if you truly believe it’s dangerous, don’t do it.