On July 17, at the main forum of the 2026 World AI Conference and High-Level Meeting on Global AI Governance, SenseTime Chairman and CEO Xu Li delivered a keynote speech titled “Boundless Innovation and Guarded Boundaries: The Inclusivity and Safety of AI Development.”
Xu Li pointed out that AI technology is accelerating, empowering every industry, and steadily moving toward general artificial intelligence—showing incredible potential for boundless innovation. But as the tech races ahead, we urgently need a standardized, systematic framework to keep things in check. It’s about balancing innovation with governance so we can crack the code on making AI truly accessible and pushing the industry forward in a healthy, sustainable way.
On the practical side, AI has already closed the loop in many fields, breaking down technical barriers and transforming from a specialized tool into something everyone can use. Take software development, for example—it used to be just for professional coders. Now, regular folks can tap into skills that were once only for senior engineers. Content generation, image and video creation—these have all been proven to work in real-world cycles, and they’re reshaping how we think about creativity.
At the same time, physical AI is evolving. Robotics, embodied intelligence, and similar technologies are moving from single-module tasks to full, autonomous long-range operations. They’re adapting to general scenarios, and the user base is shifting from pros to everyday people. The boundaries of tech application just keep expanding.
So, what does it really mean to make AI accessible to everyone? And what does it mean to use AI for good?
Xu Li said, “Every big leap in productivity tends to redefine a ton of new jobs. Sure, some existing processes will get automated, but the potential for growth is way bigger.”
Of course, a lot of AI tools create dependency—on the tools themselves, on how we use them. But a good AI product isn’t about dependency. It’s about focusing on personal growth, helping you level up, and unlocking new possibilities. When you use AI and your own skills improve, that’s a product built to last. It’s like those popular short dramas where people travel back in time. Even without modern tools, they’re still incredibly capable—because their abilities have become second nature.
He also shared a story about a visually impaired person who cared about China’s 17 million visually impaired individuals. This person used a set of tools to create educational materials tailored to their needs—because they understood what those individuals really needed. They interacted with the model entirely through voice, and AI truly extended their capabilities to new frontiers.
When AI pushes the boundaries to the point where everyone becomes a super-individual, the real future of billing lies in the price of completing tasks. Moving from token-based to task-based pricing is a whole new way of thinking about economics. And as task costs keep dropping, AI will become genuinely accessible to all.
He mentioned that along the way, a lot of people worry some jobs will disappear—especially those that are process-driven and sit between two systems. Those will definitely get optimized by AI. But the expanding boundaries will also create tons of new careers, new company structures, and new organizational forms. According to World Economic Forum data, by 2030, we might lose 90 million jobs, but we’ll gain 170 million new ones. Super-individuals will redefine job categories, fueling growth across entire industries.
So, how do we keep AI developing safely within boundaries? Xu Li broke it down into three key principles:
First, a product philosophy: don’t replace, enhance; don’t create dependency, foster personal growth.
Second, model development safety: ensure data security during pre-training and post-training, and use multi-model cross-validation during use to keep models safe in practice. It’s a new paradigm for usage.
Third, physical boundary safety: use sandboxes and limit embodied AI to specific scenarios, gradually building safe boundaries for real-world AI deployment.
He also stressed that AI governance is a global challenge—you can’t do it alone. SenseTime has worked with the United Nations to build the AI Government for Humanity Lab, contributed to UN global governance reports, and partnered with multiple countries on ethics-based governance and AI education initiatives. Only by breaking down governance silos and redefining the next phase of AI use with more partners can AI development become truly healthy.
Closing his speech, Xu Li quoted the Tao Te Ching: “What is, is for benefit. What is not, is for usefulness.” Today, we focus on the “what is”—model development, parameter changes, application breakthroughs. But the governance framework behind it is the “what is not.” Only when we prioritize both can we truly use AI well: make it accessible so everyone can use it, make it good so it’s used the right way, and make it safe so people feel confident using it.
