Bill Gates supports the establishment of a legal supervision system for artificial intelligence, wit

Bill Gates warns that AI companies should not self-regulate, urging Washington to enact security regulations

Bill Gates, co-founder of Microsoft, issued a warning that artificial intelligence (AI) companies should not be allowed to self-regulate and urged the U.S. government to enact relevant security regulations. He made this point in a September 25 interview with NBC News, the full version of which was later aired on "Meet the Press."

Gates 'warning comes as Goldman Sachs predicts that global AI investment will exceed the US$1 trillion mark in 2026. This means that new safety regulations could have a significant impact on developers, investors and related companies working to implement new security systems to enforce the rules.

"No one thinks self-regulation is enough"

When host Kristen Welker asked Congress if action was needed, Gates responded: "Absolutely." "No one thinks self-regulation is enough," he said. He believes that legislators and law enforcement agencies should enact clear laws that stipulate how AI companies monitor risks and what safeguards must be taken. He described these responsibilities as "a little burden" on developers rather than a stumbling block to progress.

他的担忧远超出了通常意义上的滥用问题。盖茨在接受NBC采访时表示，AI“确实拥有足以引发导致十亿人丧生事件的威力”，尤其是当这些先进系统落入别有用心者手中时。

行业领袖们对于安全措施应延伸到何种程度持有不同意见。Anthropic首席执行官达里奥·阿莫迪（Dario Amodei）提出了“全面放缓，甚至‘暂停’”的概念，将其视为最极端的集体减速方式。OpenAI首席执行官山姆·奥特曼（Sam Altman）也表达了类似的担忧，他认为“保持节奏的成本是值得的”，并且不应以竞争压力为借口而鲁莽行事。

然而，Meta首席执行官马克·扎克伯格（Mark Zuckerberg）持相反观点。他在接受NBC新闻采访时称：“我认为我们不需要某种形式的行业协调。”他倾向于让实验室仅在感到面临安全风险时才放慢脚步。

盖茨希望管理风险，而非 halted 开发

盖茨并非仅仅要求科技部门停止开发AI系统。他在8月发表于其博客《盖茨笔记》（Gates Notes）的一篇文章中指出，人工智能“要么成为有史以来最伟大的均衡器，要么成为最不公正的来源”。

He also said he might support a credible global AI slowdown plan, but he was not convinced that such a plan would be effective because of the strong impetus for continued development for economic and political reasons.

So his stance is not just to end new developments, but to prepare businesses, governments and individuals to respond appropriately when AI technology is applied in areas such as medical, legal, software and customer service.

Trillion-dollar market waiting for rules to be implemented

The debate on security measures is unfolding at the same time as money is pouring into the AI space, making any new rules stakes higher. Goldman Sachs predicts that global AI investment will exceed US$1 trillion by 2026, with China and the United States accounting for approximately US$581 billion, and cumulative investment since 2022 may approach US$1.8 trillion by the end of the year.

Despite the significant increase in AI investment, Microsoft points out that not everyone enjoys its advantages and benefits equally. Microsoft's AI proliferation data shows a growing gap between the global North and the global South, while the OECD emphasizes that AI capabilities are still limited to a few countries and companies.

This makes regulation a more difficult art of balancing. Better protection measures can reduce risks, but the additional cost of complying with these regulations may be unaffordable for small laboratories and less affluent countries.

As security spending and regulation accelerate, AI investment exceeds US$1 trillion

Regulatory collage is taking shape

There is growing evidence to support calls for stricter regulation. The 2026 International AI Security Report, directed by Turing Award winner Yoshu Bengio and co-written by more than 100 field experts, describes existing dangers and potential future threats posed by enhanced capabilities of AI systems.

At the same time, governments have begun to develop regulations around these risks. According to a report from TRM Labs, a new global AI regulatory "collage" is taking shape as countries adopt different approaches to AI governance, implementation and accountability.

The European Union is more active than most other countries in promoting the implementation of such goals. Under the EU AI Act, general AI providers must comply with strict obligations regarding transparency of documentation, copyright and training data. Models that use more than 10^25 FLOP computing power are considered systemic risks, which require additional model evaluation, risk management, incident reporting and cybersecurity standards to be met.

As regulation increases, so does the amount of money companies spend trying to comply. According to Gartner, spending by organizations on safeguarding AI will increase by 68.7% to nearly $4.8 billion by 2027 and is expected to reach nearly $7.7 billion by 2028.

Washington also appears to be moving towards a stronger regulatory framework. Cryptopolitan reports that Senate negotiators are studying a bipartisan initiative to establish a "duty of care" for AI developers that could allow the government to block the release of unsafe models. If approved, it would shift the protection mechanisms Gates urged from the area of voluntary obligations to the area of legally binding responsibilities.