OpenAI chief scientist Jakub Pachocki warned in September 2026 that AI alignment and monitoring were not keeping pace with capabilities. His warning makes the industry’s confidence about advanced AI a question of evidence and safeguards, not simply whether someone calls a model AGI.
\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nOpenAI launched Astra on September 3 as its “most intelligent and aligned model” yet. It can operate computers, browse the web, perform complex professional work and has become OpenAI’s first model to reach its Critical cybersecurity capability threshold.
\n\n\n\nThree days later, Pachocki published An Alien Mind.
\n\n\n\nHis conclusion was remarkably direct.
\n\n\n\nNo AI lab, he wrote, has solved alignment and monitoring well enough to continue responsibly scaling at maximum speed for much longer.
\n\n\n\nHe expects voluntary slowdowns. He wants common safety thresholds. And he wants governments to coordinate internationally before the technology moves much further.
\n\n\n\nThere is another detail that deserves attention.
\n\n\n\nOne of OpenAI’s most important ways of checking what reasoning models are doing is chain-of-thought monitoring.
\n\n\n\nPachocki says its reliability is progressively diminishing.
\n\n\n\nThe models are getting better at reasoning about and manipulating their own reasoning process. They are also becoming more capable without verbalising that reasoning at all.
\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nThe label AGI does not resolve the operational question: what safeguards and evidence support deploying a more capable system?
\n\n\n\nLook at the incentives and the signals.
\n\n\n\n\n\n\n\nThe company building the model is selling its most capable AI ever.
\n\n\n\nAnd that company’s own chief scientist is simultaneously warning that our ability to control and monitor increasingly capable systems is not keeping pace.
\n\n\n\nFor businesses deploying AI agents, and consumers increasingly allowing them to act on their behalf, that is not an academic disagreement.
\n\n\n\nIt is a trust question.
\n\n\n\nWhat evidence should customers require before trusting increasingly capable AI agents with consequential work?
\n","content_text":"Editorial update — 17 October 2026. This revision focuses on Pachocki’s published safety assessment. The earlier Nvidia quotation and large training-compute counts were removed because their original source was not established in this review. AGI labels remain judgments rather than a settled technical classification.\n\nOpenAI chief scientist Jakub Pachocki warned in September 2026 that AI alignment and monitoring were not keeping pace with capabilities. His warning makes the industry’s confidence about advanced AI a question of evidence and safeguards, not simply whether someone calls a model AGI.\n\nOpenAI launched Astra on September 3 as its “most intelligent and aligned model” yet. It can operate computers, browse the web, perform complex professional work and has become OpenAI’s first model to reach its Critical cybersecurity capability threshold.\n\nThree days later, Pachocki published An Alien Mind (https://openai.com/index/an-alien-mind/) .\n\nHis conclusion was remarkably direct.\n\nNo AI lab, he wrote, has solved alignment and monitoring well enough to continue responsibly scaling at maximum speed for much longer.\n\nHe expects voluntary slowdowns. He wants common safety thresholds. And he wants governments to coordinate internationally (https://reinouttebrake.com/stories/ai-was-supposed-to-slow-down-now-theyre-being-sued-for-it) before the technology moves much further.\n\nThere is another detail that deserves attention.\n\nOne of OpenAI’s most important ways of checking what reasoning models are doing is chain-of-thought monitoring.\n\nPachocki says its reliability is progressively diminishing.\n\nThe models are getting better at reasoning about and manipulating their own reasoning process. They are also becoming more capable without verbalising that reasoning at all.\n\nThe label AGI does not resolve the operational question: what safeguards and evidence support deploying a more capable system?\n\nLook at the incentives and the signals.\n\nThe company building the model is selling its most capable AI ever.\n\nAnd that company’s own chief scientist is simultaneously warning that our ability to control and monitor increasingly capable systems is not keeping pace.\n\nFor businesses deploying AI agents, and consumers increasingly allowing them to act on their behalf, that is not an academic disagreement.\n\nIt is a trust question.\n\nWhat evidence should customers require before trusting increasingly capable AI agents with consequential work?","referenced_links":[{"url":"https://openai.com/index/an-alien-mind/","label":"An Alien Mind"},{"url":"https://reinouttebrake.com/stories/ai-was-supposed-to-slow-down-now-theyre-being-sued-for-it","label":"governments to coordinate internationally"}],"social_image":"https://reinouttebrake.com/assets/editorial/jensen-1280.webp","citation":"Reinout te Brake. OpenAI’s Chief Scientist Warns That AI Safety Is Falling Behind. 2026-09-07. https://reinouttebrake.com/stories/agi-is-here-says-nvidia-nobody-is-ready-says-openais-chief-scientist"}