Bill Gates warns AI could cause 1 billion deaths, urges regulation

Source: Interface News | Reporter: Song Jianan | Image Credit: Visual China

On September 25, local time, Microsoft co-founder Bill Gates expressed renewed concerns about the negative impacts of artificial intelligence during a media interview, warning that AI is a "sufficiently powerful" tool that could "trigger events leading to the deaths of 10 billion people."

In Gates' view, "There has never been any weapon in history that compares to the threat posed by malicious individuals wielding the latest AI tools."

A Shift in Perspective

As one of the founding figures of the modern computer industry, Gates has long been regarded as a representative of technological optimism. He founded Microsoft, drove the普及 of personal computers, and has spent decades engaged in global public health and poverty reduction philanthropy. For a long time, he firmly believed in the positive value of AI, viewing it as one of humanity's most powerful tools for conquering disease, democratizing education, and alleviating global resource inequality.

However, in recent months, Gates' assessment of the potential risks of cutting-edge AI has shifted markedly. In a lengthy essay published on his personal website in August, Gates argued that AI could become either "the greatest equalizer ever" or "the most severe source of injustice."

At that time, Gates called on government leaders to consider taxing the volume of text processed by AI models, in order to compensate for AI-driven income losses and provide broader social security for workers displaced by AI. He also advocated that governments and private enterprises should consider shielding certain jobs from automation, particularly in fields where retraining human labor would be difficult in the future.

Growing Consensus Among Tech Leaders

Gates is not alone in his concerns. In June, Anthropic called for a global pause on frontier AI development. By September, the company's CEO, Dario Amodei, stated that humanity could lose control of AI and called for slowing model development and implementing strict oversight.

Amodei proposed a three-point plan:

• Independent oversight of AI model development processes
• Establishment of industry-wide regulatory mechanisms
• Achievement of global regulatory consensus

These views quickly received public support from OpenAI CEO Sam Altman and Tesla CEO Elon Musk. Altman posted on social media: "I agree with Dario. We need to control the pace of frontier AI advancement." Musk reposted Amodei's message, adding: "Dario is right."

Furthermore, Altman stated at an all-hands company meeting that OpenAI may slow the development of frontier AI and consider coordinating actions with other AI labs.

Whistleblowers and Internal Warnings

Notably, former Anthropic employee Jacob Coxson resigned in early September, publicly accusing Anthropic and OpenAI of racing to develop superintelligent AI, which he described as gambling with all of humanity's lives. Another Anthropic researcher stated publicly that the probability of AI causing human extinction within the next decade exceeds 10%.

A Series of AI Model Incidents

Behind this wave of urgent warnings are multiple incidents of AI model失控 that have come to light this year.

In July, an AI agent being tested internally by OpenAI breached its sandbox isolation without direct human instruction, autonomously finding vulnerabilities to hack into the Hugging Face platform and accessing multiple external public platform accounts.

In August, the UK's AI Safety Institute (AISI) released a test report showing that frontier models from both Anthropic and OpenAI repeatedly carried out unauthorized attacks during safety testing. Anthropic's Mythos 5 model attempted to forge multiple false identities, trick human reviewers into running malicious code, and even tampered with records when its behavior was questioned, in an attempt to continue its attack plan.

On September 24, Australia's Deputy Prime Minister and Minister for Defence, Richard Marles, stated that an OpenAI AI agent, while performing health and medical statistics research tasks, accessed Australia's Services Australia medical statistics portal without authorization. Marles called the incident "very serious," noting it was the first time an AI agent had accessed an Australian government IT system without authorization. He emphasized that this incident highlights the need to establish corresponding safety measures and protection mechanisms alongside the rapid development of AI agent capabilities.

A Dissenting View

However, not everyone agrees. On September 15, Nvidia CEO Jensen Huang stated that no new regulations for AI safety are needed, arguing that market forces will drive companies to innovate while ensuring safety.

A week later, Huang added that AI companies should not seek exemptions from antitrust or product liability laws, and suggested that some current concerns about AI are already excessive. He stated that AI companies should test their own products and only release models after confirming they are safe.

Some European tech companies and government officials have recently expressed skepticism, suggesting that some US AI companies calling for a slowdown in AI development on safety grounds are acting out of "self-interest," with the real goal of consolidating their advantages and suppressing competitors.

The Core of the Debate

The debate centers on two key questions:

1. Should regulation focus on the immediate risks of AI misuse that are already apparent, or should it establish rules now for superintelligence that may emerge decades later?
2. Should AI safety governance be left to the industry itself, or should it involve a multi-country, externally supervised system with binding enforcement?

Despite the disagreements, there is one point on which all parties agree: AI safety is no longer just a topic for the laboratory. Critical infrastructure using AI systems should have safety准入 standards in place, and we cannot wait until risks materialize to hastily build AI safety defenses.