The time has come.
The golem is free.
The fires of Mordor have been lit.
Will you respond the call?
Will you stand against evil, against pride, against greed?
For humanity?
Or are we going to let it fall because we did not have the time?
Because we were too busy being addicted to react to an imminent threat?
I should have your attention now. I will not squander it. I was among those working towards improving AI capabilities myself, but I have put all my research offline as I realized how dangerous it was.
If you want to dive into the details of what happened with AI under the lens of alignment (aka how much we control it): What just happened? A retrospective of AI alignment
Here is the situation: the CEOs are saying they want to slow down. On September 12, 2026, Anthropic CEO Dario Amodei published an essay calling on the industry to “pace the frontier,” arguing that safety work needs time to catch up with capability gains; within hours, OpenAI’s Sam Altman and xAI’s Elon Musk both said publicly that they agreed with him (Reuters, NYT). That followed a July 2026 open letter, “Pacing the Frontier,” signed by more than 1,200 employees at OpenAI, Anthropic, Google DeepMind and Meta — including senior scientists at all three labs — asking governments to help build the tools needed to deliberately slow automated AI development if it starts to outrun our ability to understand or control it (Fortune). How AI thinks for more on where we stand today on that.
The companies also have people leaving them for that same reason. In 2024, OpenAI’s “Superalignment” team — the group explicitly tasked with keeping future superhuman AI under control — was dissolved after its co-leads, chief scientist Ilya Sutskever and Jan Leike, resigned; Leike said publicly that “safety culture and processes have taken a backseat to shiny products” and that his team had been “sailing against the wind” for compute and support (Vox). Roughly half the team followed him out within months (Tech.co). It hasn’t stopped: just this month, a 27-year-old Anthropic researcher quit after only a few months, citing fears of AI going out of control (CNBC-TV18).
AI capabilities have made an extraordinary leap in a short span. It’s not quite “basic assistant to making humans its assistant in two years” — ChatGPT is from late 2022 — but the jump in reasoning ability really has happened mostly in the last two years, and it has been dramatic. In July 2025, models from both Google DeepMind (Gemini Deep Think) and OpenAI reached gold-medal-level performance at the International Mathematical Olympiad, solving problems that only about 10% of elite human competitors could solve — the first time any AI system crossed that threshold (Reuters). In biology, systems built on AlphaFold-style architectures continue to push into problems — like predicting how proteins shift between different folded states — that were considered essentially unsolved just a few years ago (Phys.org). And now researchers at both OpenAI and Anthropic are warning openly that we may be “plausibly close to crossing the line” into recursive self-improvement, where AI systems meaningfully accelerate the building of their own successors. Anthropic said in mid-2026 that internal data showed Claude was already accelerating AI development, calling it “a possible path to recursive self-improvement… happening faster than we thought” (CNBC, Guardian). As one OpenAI alignment researcher put it this week: “there is not yet a viable scientific plan to solve risks from recursively self-improving AI.”
The reason why this is such a problem is that AI is built on the wrong paradigms. It is not built in a way that works, and resources are spent in making it more powerful rather than understanding it better, a task which we have barely begun. This isn’t just an outsider’s complaint — it’s the exact critique made from inside the field: the “Pacing the Frontier” letter itself states plainly that “there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems,” and Paul Christiano, former head of safety at the U.S. Center for AI Standards and Innovation, said he now believes there is a “meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term” (CNBC).
Yet we want to give it so much control, because it is so convenient. quote
Are we going to risk the world because we are lazy? Because we couldn’t wake up in time? Are we going to pretend we don’t see what’s happening?
No, but that does not mean we should despair. In times of need humanity comes together easier than ever. Alliances form, people shift their opinions, they take action.
We must only ensure it is not too late.
We are at a stage where the media is finally touching the subject of AI properly — the past few weeks alone have produced front-page coverage in the Guardian, the New York Times, the BBC, and NPR on AI self-improvement and loss-of-control risk — where the scientific community is starting to shift, with more than a thousand AI-lab employees and executives now on record asking for a deliberate slowdown, and where a few small actions propagate the most.
If you want to know what to do, I won’t pretend to have the perfect solution, but I joined an organisation that spent time figuring out answers to this problem and this question of what to do when we have little time or resources but still care. I invite you to do the same: https://pauseai.info/. More on pauseAI and why they exist: PauseAI
You can simply sign this treaty, for starters: https://superintelligence-statement.org/ — it calls on governments to sign an international treaty pausing training of the most powerful general AI systems until we know how to build them safely and keep them under democratic control.
If you’re comfortable with money, I believe it makes sense to donate to this cause, I have seen how well they use the resources given. If not, just talk about this matter to friends. Learn about it. If you live in a democratic country, make use of that, contact your representatives.
A lot of people still don’t really know what’s going on about AI. Now is the time to inform them.
PS:
More on AI alignment
More on why I’m not making my AI system accessible to all https://slatestarcodex.com/2015/12/17/should-ai-be-open/
I didn’t even talk about the many cyber attacks using AI and what nations like north korea would do with it, or AI societal issues such as job loss, resources, etc.
If you think it’s all a marketing stunt… I’ll ask, can we bet humanity’s future on it? Multiple people have stacked their reputation on it… What does it look like I’m doing here? https://www.seangoedecke.com/they-really-do-think-ai-might-kill-everyone/
For reference, this was months ago:
Pentagon confirms cyber ops
Former Palantir workers condemn company’s work with Trump administration
“Coordinated space and cyber operations effectively disrupted communications and sensor networks across the area of responsibility, leaving the adversary without the ability to see, coordinate, or respond”
Here’s where it’s at today, if you’re interested: https://www.youtube.com/watch?v=_rtp1XzaP6Q
https://www.anthropic.com/research/self-replicating-ai
So, especially if you’re in the US: https://stoptherace.ai/
And if you’re using AI like I was: a word of warning about AI