Artificial intelligence is advancing at a dizzying speed. Like many new technologies, it offers significant benefits but also poses safety risks. Recognizing the potential dangers, leading researchers from Google DeepMind, OpenAI, Meta, Anthropic and a coalition of companies and nonprofit groups have come together to call for more to be done to monitor how AI systems “think.”
In a joint paper published earlier this week and endorsed by prominent industry figures, including Geoffrey Hinton (widely regarded as the “godfather of AI”) and OpenAI co-founder Ilya Sutskever, the scientists argue that a brief window to monitor AI reasoning may soon close.
Improving AI monitoring
They are calling for more monitoring of chains-of-thought (CoTs), a technique that enables AI models to solve complex challenges by breaking them down into smaller steps, much like humans work through complicated tasks, such as a tricky math problem.
CoTs are key features of advanced AI models, including DeepSeek R1 and Language Learning Models (LLMs). However, as AI systems become more advanced, interpreting their decision-making processes will become even more challenging. This is a concern because existing AI oversight methods are imperfect and can miss misbehavior.
In the paper, the scientists have highlighted how CoT monitoring has already proved its worth by detecting examples of AI misbehavior, such as when models act in a misaligned way “by exploiting flaws in their reward functions during training” or “manipulating data to achieve an outcome.”
The scientists believe that better monitoring of CoTs could be a valuable way to keep AI agents under control as they become more capable.
“Chain of thought monitoring presents a valuable addition to safety measures for frontier AI, offering a rare glimpse into how AI agents make decisions,” said the researchers in their paper. “Yet, there is no guarantee that the current degree of visibility will persist. We encourage the research community and frontier AI developers to make the best use of CoT monitorability and study how it can be preserved.”
One key request from the researchers is for AI developers to study what makes CoTs monitorable. In other words, how can we better understand how AI models arrive at their answers? They also want developers to study how CoT monitorability could be included as a safety measure.
The joint paper marks a rare moment of unity between fiercely competitive tech giants, highlighting just how concerned they are about safety. As AI systems become more powerful and integrated into society, ensuring their safety has never been more important or urgent.
Written for you by our author Paul Arnold,
edited by Gaby Clark, and fact-checked and reviewed by Andrew Zinin—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive.
If this reporting matters to you,
please consider a donation (especially monthly).
You’ll get an ad-free account as a thank-you.
More information:
Tomek Korbak et al, Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety, arXiv (2025). DOI: 10.48550/arxiv.2507.11473
arXiv
© 2025 Science X Network
Citation:
Tech giants warn window to monitor AI reasoning is closing, urge action (2025, July 17)
retrieved 17 July 2025
from https://techxplore.com/news/2025-07-tech-giants-window-ai-urge.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
