Meta founder and CEO Mark Zuckerberg entered the national debate over artificial intelligence regulation this week with a basic question: What happens if the most powerful AI systems are controlled by too few people?
In an essay published Aug. 10 on Meta’s official newsroom, “The Future is for Everyone,” Zuckerberg outlined his case for what he calls personal superintelligence, advanced AI designed to work on behalf of individual users. He argues those systems should be broadly available rather than concentrated among governments, corporations or a small group of experts.
“The defining questions of our age are who will have access to superintelligence and what will we direct it towards,” Zuckerberg wrote.
The essay arrives as AI companies, researchers and lawmakers debate how much oversight should exist as models become more capable. The dispute is increasingly about where regulation should begin and who should have authority to draw the line.
According to the 2026 International AI Safety Report, general-purpose AI systems are becoming more capable in areas including programming and scientific research, while their abilities remain uneven. The report describes an “evidence dilemma” for policymakers: acting before risks are well understood can create ineffective restrictions, while waiting for definitive evidence may leave governments responding after serious harm occurs.
Zuckerberg’s argument starts with the distribution of power. He says concentrating advanced intelligence among a limited number of institutions can create its own safety problem and favors a model in which individuals, businesses and governments have access to comparable tools.
Cybersecurity is one example. Zuckerberg argues that if only one actor had access to superior cybersecurity capabilities, that actor could exploit systems others could not adequately defend. If defenders had comparable tools, they could use those capabilities to find vulnerabilities and strengthen systems.
Meta’s vision goes beyond an AI assistant that answers questions. Zuckerberg describes personal agents that could help people learn, work, start businesses and contribute to scientific research.
He also addresses cybersecurity, biological and chemical threats, job disruption, surveillance and the possibility of humans losing control of advanced systems. His response generally favors stronger defenses, government cooperation and targeted safeguards rather than broad restrictions on individual access.
On biological and chemical threats, Zuckerberg says developers should reduce dangerous capabilities in their models while governments focus on controlling the physical production and distribution of harmful materials.
Other AI developers place the line earlier.
Anthropic, the company behind Claude, says in its Advanced AI Framework that frontier developers should test powerful models, disclose safety findings, submit systems to independent evaluation and maintain security programs. For models presenting a significant risk of catastrophic harm, Anthropic says government should have authority to “block or deter” deployment.
The company also says that authority should include safeguards against government overreach.
The distinction is central. Zuckerberg warns that restricting powerful AI can concentrate power among the institutions allowed to possess it. Anthropic argues that some capabilities may become consequential enough that waiting for misuse is not sufficient.
OpenAI has also supported a larger federal role. In a June blueprint on frontier AI governance, the company proposed a national framework, a strengthened federal institution focused on frontier AI safety and broader government preparation for national security and public safety risks.
According to the International AI Safety Report, AI systems are already being used in fraud, cybercrime and manipulation. At the same time, many capabilities raising concern are dual-use. A model that identifies software vulnerabilities can help an attacker or help defenders repair weaknesses. AI can also make complex biological and chemical information easier to understand while supporting legitimate research.
Illinois has placed much of the regulatory responsibility on companies building frontier systems.
Gov. JB Pritzker signed the Artificial Intelligence Safety Measures Act into law July 6. According to the Illinois General Assembly, it requires large frontier AI developers to maintain frameworks addressing catastrophic risks, cybersecurity and governance, along with transparency, independent audit and critical safety incident reporting requirements. The law takes effect Jan. 1, 2027.
State Sen. Mary Edly-Allen, the measure’s chief Senate sponsor, said the legislation was intended to establish safeguards without stopping AI development.
“Power has to be matched with wisdom, transparency and responsibility,” Edly-Allen said. “This law is not about stopping innovation.”
The law does not broadly determine what an individual may ask an AI system. Instead, it increases obligations on certain developers as their systems become more capable.
The emerging debate leaves a basic question: Should regulation focus on the information an AI system provides, the actions it can perform, the company that develops it, the person who uses it or the illegal conduct that may follow?
Technology companies understand the systems they build but also have financial interests in deploying them. Governments can enforce limits, but authority over access to advanced technology raises questions about surveillance, political power and control over information.
The line may also depend on capability. An AI system helping a student understand calculus presents different concerns from one identifying vulnerabilities in critical infrastructure. A system assisting scientists with protein research presents different risks from one that lowers barriers to harmful biological work. The International AI Safety Report similarly distinguishes between beneficial and harmful uses of the same underlying capabilities.
Zuckerberg’s position is that advanced intelligence should be broadly distributed while governments and defenders are strengthened to counter misuse. Anthropic argues that some systems may warrant intervention before deployment. OpenAI has called for federal oversight, while Illinois has begun placing new requirements on developers.
The debate is no longer only about whether AI should be regulated. It is increasingly about where the line belongs and who gets to draw it.