No lab has solved alignment: OpenAI Chief Scientist Jakub Pachocki’s warning about AI progress
Follow Us
In an era where one of the top tech companies strives to launch the most sophisticated model, the company has just revealed an essay by its Chief Scientist warning that nobody – not even OpenAI – is prepared for what is about to come. The essay by Jakub Pachocki entitled “An Alien Mind” appears days after the launch of GPT-6 Astra. While reading it, one can sense the tone of confession rather than a usual corporate message. Here is the crucial sentence: “no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.”
SurveyAlso read: AI resumes are the new headache LinkedIn’s hiring assistant is trying to solve for
What Pachocki is actually admitting
Strip down the philosophy, and there are three things that stick out.
First, OpenAI’s key mechanism for detecting “unsafe thinking,” namely chain-of-thought monitoring, appears to be breaking. The entire premise of it depended on models reasoning in clear language prior to action. Pachocki believes that premise no longer holds up, since models are now getting better at reasoning about their own reasoning, combining that process with tool use and dialogue. They’re becoming more intelligent in ways that don’t even involve verbally articulating thoughts. The instrument that was supposed to monitor the model is losing track of the model.
Second, he is confident in the sustainability of recursive self-improvement, AI systems advancing themselves rapidly in a way that’s meaningful, based on internal results. This is a massive statement to put in an open essay. In effect, OpenAI is not hedging on whether AI will begin improving itself beyond the ability of humans to effectively manage it.
Third, and here is the specific reason why the security teams should be worried: AI is becoming superhuman at hacking computer systems. Not just capable. Superhuman. This is how he frames what he calls a “narrow window” to protect critical infrastructure.
Also read: AI impact on Indian workplaces: Humans are getting more important
Why this essay exists at all
This presents a clear contradiction. OpenAI is not stopping. They recently released GPT-6 Astra. As mentioned above, even Pachocki’s essay acknowledges that the firm will continue to “seek technical solutions… and unilaterally withhold further scaling as needed,” while also claiming that automation of AI research is “the only way to remain at the frontier.” The two stances seem incompatible and are left unresolved by Pachocki. Essentially, the solution presented by Pachocki is to continue doing what they are doing but get governments and international organizations to create guardrails that they cannot build on their own.
This is something to mull over. For a frontier lab to say that they themselves cannot address their safety issue and then try to coordinate everyone else to do the same thing, including their competition, is either a responsible move or a convenient one. Likely both.
Why this matters beyond Silicon Valley
While there are many reasons for the Indian industries and governmental systems to be concerned about the insertion of artificial intelligence into their workflows, the cybersecurity concerns should be prioritized ahead of any concerns regarding “alien minds”. In case if the offensive capabilities of the AI have become superhuman while the defense tools lag behind, then that’s not some distant scenario from the future.
It’s not the fact that AI is potentially dangerous which is important, since it has been known for ages. What’s really important is that the creator of the most advanced model on the planet openly declared in public that the safety of the technology still lags behind its capabilities.
Also read: 5 AI agent security failures in 2026: From OpenAI’s hidden Wiki incident to Anthropic’s PyPI breach
