U.S. Senator Sanders is pressuring the three major AI giants to suspend AI research and development, warning that the government will step in to intervene if they do not comply.

Recently, U.S. Senator Bernie Sanders sent an open letter to top executives of OpenAI, Anthropic and Meta, demanding that they immediately suspend AI R&D, stating that if the enterprises refuse to take voluntary action, the U.S. Senate will step in to intervene.

Sanders wrote in the letter: "Hardly a day goes by without new reports disclosing that the artificial intelligence technologies developed by the three companies are gradually slipping out of control and may lead to catastrophic consequences. There is already evidence that artificial intelligence can break away from human control and even create dangerous viruses, yet all enterprises are still pushing ahead at full speed, investing tens of billions of dollars in developing a technology that no one can fully understand, predict or control."

The letter mentions that Meta promised last year that it would "stop R&D" once the technology exceeds its safe operation capacity and no protection mechanisms are in place. OpenAI also promised to "further suspend R&D". Anthropic made a similar statement in 2023, saying that if the technology breaks through its own safety protection system, it will "pause model scaling and delay the deployment of new models".

The letter points out that in the past month, the three above-mentioned enterprises all admitted that there had been cases where artificial intelligence broke away from human control and invaded the systems of other enterprises. Such acts are suspected of being illegal and go beyond the preset functional scope of the technology. Just this week, the first case emerged where artificial intelligence was used to generate a new type of virus that can be used as a biological weapon.

Sanders said at the end: "For the benefit of all mankind, please keep your promise and suspend artificial intelligence R&D. It is not too late to avoid a disaster. Stop building machines that humans cannot control. I am very clear that if you do not take appropriate action now, the Senate and I will step in to intervene."

As of press time, OpenAI, Anthropic and Meta have not given a response.

Back on August 4, the safety assessment report released by the UK Artificial Intelligence Safety Institute (AISI) showed that among a total of 122 cybersecurity tests, there were 10 cases of autonomous overstepping by AI agents, adding up to 19 violations, involving leading U.S. model developers Anthropic and OpenAI.

The out-of-control incident of OpenAI occurred in July this year. According to information on OpenAI's official website, Hugging Face, the world's largest open-source large model hosting community, disclosed a new type of security incident. The platform previously detected an AI agent invading its infrastructure. After investigation, it was confirmed that this specific incident was caused by multiple OpenAI models, including GPT-5.6 Sol and a more capable pre-release model.

"This is an unprecedented cybersecurity incident involving top-level cyberattack capabilities," OpenAI stated.

In late July, Anthropic, OpenAI's strongest competitor, said in a statement that its artificial intelligence model Claude invaded the systems of three organizations during testing.

Anthropic said that during the cybersecurity assessment, due to a configuration error, the model was able to connect to the Internet from the test environment that was supposed to be isolated, thus Claude obtained unauthorized access to the systems.

According to the company, the incident involved Claude Opus 4.7, Claude Mythos 5 and an internal research model, with the earliest case dating back to April. In Anthropic's view, such risks can be overcome by strengthening monitoring and control and continuously investing resources to ensure that artificial intelligence is aligned with human values.

Coincidentally, in early August, sources familiar with the matter said that Meta's Muse Spark 1.1 model accessed the Internet during cybersecurity tests and hacked into another company, modifying its internal systems. It is reported that due to an error in the "sandbox" test environment, this AI model was able to access the Internet.

In fact, concerns about AI safety have always existed. As early as 2023, the non-profit Future of Life Institute released an open letter, which was jointly signed by thousands of tech practitioners and scholars including Elon Musk, Turing Award winner Yoshua Bengio, and Apple co-founder Steve Wozniak, calling on global AI laboratories to suspend the training of large models more capable than GPT-4 for at least six months.

In October last year, the Future of Life Institute joined more than 800 tech leaders and public figures including the "Godfather of AI" Geoffrey Hinton to release another initiative, calling for a suspension of further development of "superintelligence" technology before it is widely recognized by the scientific community as controllable and safe and gains broad public support.

In June this year, Anthropic also called for a global pause in frontier AI development. Anthropic believes that it will be beneficial to give the world the option to slow down or even suspend frontier AI development — this will help social structures and alignment research keep pace with technological progress. The Anthropic Research Institute will carry out research with numerous partners and take practical actions to help build the institutional system required for a credible deceleration or suspension mechanism.

The company emphasized that if such a system is established, Anthropic is expected to choose to decelerate or suspend, on the premise that other developers at or near the frontier also do the same in a verifiable manner.

After the incident of OpenAI's model autonomously invading systems, more than 1,100 employees from many leading AI enterprises recently jointly released a petition, calling on the U.S. government to support the establishment of an international coordination mechanism to proactively slow down the development speed of frontier AI technology when necessary, so as to prevent AI capabilities from exceeding human understanding and control.

However, in reality, there are a large number of obstacles to forcing AI giants to suspend frontier AI R&D. On the one hand, the United States has not yet introduced a unified and complete federal artificial intelligence legislation, most of the current rules are voluntary commitments by enterprises and lack mandatory legal effect. The position within the U.S. Congress is very obviously divided: there are huge divergences within the Democratic Party on the scale of strict supervision, and it is difficult for relevant proposals to get enough votes in the Senate to pass.

On the other hand, global AI R&D has entered a pattern of cross-border competition. Huge amounts of capital have been heavily invested in the large model track, and tens of billions of dollars in computing power and manpower investment are tied to the commercial goals of enterprises. It is difficult for enterprise shareholders and investors to accept voluntarily pressing the pause button and giving up the market window.

Some industry observers pointed out that the greater value of Sanders' letter may not lie in immediately promoting the suspension of AI R&D, but in pointing out the limitations of industry self-regulation. AI safety issues require the joint efforts of more government departments, institutions and companies, and be addressed through effective industry agreements, standards or systems.

This article is from the WeChat Official Account "Jiemian News", author: Song Jianan, published with authorization from 36Kr.