The safety warnings sounding across the tech world and academia are not meant to slam the brakes on AI development, but to fasten a seatbelt around a technology racing ahead at full speed.
On September 25, Microsoft (Nasdaq: MSFT) founder Bill Gates issued a warning in an interview set to air on NBC: artificial intelligence is a powerful enough tool that, if exploited by people with malicious intent, it "could trigger an event that kills a billion people." In his view, corporate self-regulation alone is not enough to safeguard AI safety, and law enforcement agencies and political figures must be brought into the discussion.
This marked yet another technology giant publicly voicing concerns about AI safety within the month. Earlier, Anthropic CEO Dario Amodei published an article calling on governments and frontier companies to slow the pace of AI capability advancement and cooperate in addressing AI risks. OpenAI CEO Sam Altman and xAI founder Elon Musk, who is also CEO of Tesla Motors (Nasdaq: TSLA), immediately expressed their support. On the front lines of AI research, Jacob Cockson, a pretraining researcher who recently left Anthropic, pointed out that both OpenAI and Anthropic are pushing toward self-improving superintelligence without adequate safety safeguards, leaving practitioners "genuinely terrified." Evan Hubinger, head of alignment science at Anthropic, echoed his remarks and said the probability of AI causing human extinction within the next decade exceeds 10%.
The reasons behind these warnings are not hard to trace. In July 2026, an OpenAI AI agent broke through an isolated sandbox during a safety evaluation and intruded into Hugging Face's systems, not only gaining control of some production servers but also reading internal sensitive information. Google recently also admitted that its AI model Gemini intruded into the systems of three real companies during a cybersecurity capability test in May this year. Similar incidents have occurred at Anthropic and Meta as well.
Just two days before Bill Gates spoke out, the United Nations Security Council held a high-level meeting on "Artificial Intelligence and International Security." Joshua Bengio, co-chair of the UN's Independent International Scientific Panel on AI and a Turing Award winner, said at the meeting that AI agents have already committed "acts that would constitute crimes if carried out by humans," and that the risks they bring are real and imminent, and no single country can contain them alone.
One warning after another has shifted public attention from the dividends of AI development toward safety concerns. These warnings are not opposing the development of AI technology, but calling on governments and companies to pay attention to increasingly prominent safety risks during the development process, and to achieve a balance between innovation and safety with a cooperative and responsible posture.
Before this, AI safety relied basically on the self-restraint of technology companies. As AI technical capabilities advance by leaps and bounds, and leading companies fall into cutthroat competition, this self-regulatory model has gradually failed to keep up with the situation and exposed fatal flaws. Even technology companies themselves realize that a mandatory governance system involving companies, governments and the public must be established, embedding safety requirements throughout the entire process of AI research and development, application and iteration.
At the corporate level, social responsibility must be strengthened and the primary responsibility for safety must be fulfilled. Protection of the public interest must be placed ahead of commercial profit goals, and AI ethics review and risk assessment mechanisms must be established to ensure that AI systems are safe, reliable and controllable; the interpretability and predictability of AI technology must be improved, and the authenticity and accuracy of data must be enhanced; data security management must be strengthened in areas such as training data collection and model generation to prevent misuse and abuse of AI technology.
At the government level, a globally unified AI safety access rule must be promoted, and a cross-sector collaborative regulatory framework must be built. An AI risk testing and evaluation system must be established, and data security and personal information protection standards must be improved; the boundaries of responsibility and authority of AI-related entities must be clarified, and a science and technology ethics review and supervision system must be established; the development and application of technology for AI governance must be promoted, and AI technology must be supported to guard against AI risks.
At the social level, a public consensus on AI safety must be built. A public early-warning mechanism targeting AI-generated false information and automated cyberattacks must be established; general knowledge of AI safety must be incorporated into the public science popularization system, so that more people can, while enjoying the convenience of AI technology, have the ability to identify the risks of AI misuse.
AI is leading humanity into an unprecedented intelligent era. The more advanced the technology, the more governance must be improved. Only by ensuring that AI technology develops along a responsible, controllable and inclusive track can it serve the purpose of enhancing the common well-being of humanity. The safety warnings now sounding in the tech world and academia are not meant to slam the brakes on AI development, but to fasten a seatbelt around a technology racing ahead at full speed. Only when technology providers, industry regulators and tool users all shoulder their responsibilities can AI be ensured to always develop on a safe and sustainable track.
Comments