AI is beginning to move beyond the boundaries of the systems that created it.
After years of investment and experimentation, enterprises are under pressure to turn AI into real productivity and economic value. That is pushing them toward a new generation of AI agents: software that does not simply respond to commands but can act independently.
Read More on Opinion
The opportunity is enormous. As these autonomous systems become more capable and interconnected, entirely new categories of services, automation and economic growth will emerge.
But as agents gain more autonomy, businesses and governments will need more control over what those agents can do, where they can operate and who is responsible when they act. Without that foundation, security and compliance risks compound quickly, especially as autonomous software begins moving across systems without a clear chain of responsibility.
We’ve already begun to see early evidence of those risks. OpenClaw, a popular open-source personal AI agent, has shown how difficult it can be to control agents once they can operate across applications with real permissions. At the same time, Anthropic’s Claude Mythos model, shared with select partners through Project Glasswing, shows how quickly frontier models are becoming powerful enough to identify vulnerabilities in critical software. Agents are likely to become more autonomous, and models more capable. The question is whether the standards by which they are governed can keep pace.
This is where the deeper risk emerges: that the answers to these problems are built only inside fragmented local or proprietary systems. Companies will rely on their existing systems to handle agent identity and permissions because they have to. But if every platform and enterprise creates its own private rules, the result may solve local problems while making AI harder to trust across companies, markets and borders.
The challenge is not only to control agents inside a single environment, but to make them trustworthy across many environments. That is what turns a useful technology into shared economic infrastructure.
History shows that this kind of growth only happens through open, shared standards rather than in fragmented, closed networks.
Lessons from the Early Internet
Before the internet became the backbone of the global economy, it was a collection of independent networks. They were isolated, and interoperability depended on prior agreements rather than shared standards, leading to fragmentation.
The shift came with the adoption of open, shared standards, including the Domain Name System (DNS). DNS created a common foundation that allowed systems to interoperate globally without requiring a central authority to control them. That neutral foundation did more than connect networks. It created the conditions for competition, innovation and market expansion by allowing independent systems to work together on consistent terms.
AI is now approaching a similar moment.
While today’s AI systems are advancing rapidly, the infrastructure needed to support trusted accountability across organizational boundaries does not exist yet.
This is one of the foundational challenges underlying the agent economy.
An AI agent may have one identity inside a cloud platform, another inside an enterprise environment and different credentials across the tools and systems it interacts with. But there is no durable ownership record that persists across the ecosystem as models, credentials and tools change over time.
Without that foundation, trust becomes fragmented. Open standards can help solve this by creating a shared framework for accountability that works across systems, not just inside them.
Open ecosystems do not emerge by accident. The internet economy flourished because no single company controlled the underlying standards that allowed systems to connect. Builders could create entirely new businesses and capabilities on top of a common foundation. AI should evolve the same way.
If a shared, neutral foundation for accountability does not emerge, AI will fragment across proprietary and closed systems where trust exists only within individual platforms rather than across systems. Organizations will struggle to determine who is responsible for an agent once it crosses platforms or organizational boundaries. Traceability and accountability will break down across systems. Competition will narrow as a small number of platforms define the rules for how agents operate, who can participate and how trust is established.
We have seen this before. Open infrastructure transformed the internet from a collection of isolated networks into a global engine for innovation and economic growth.
AI has the potential to become an even larger economic accelerator. But realizing that potential requires more than increasingly powerful models. It depends on establishing the foundations that allow these systems to work together across boundaries.
The window to establish those foundations is now. Once embedded, the structure of these systems becomes far more difficult to change, and the opportunity to shape a more open, interconnected ecosystem narrows accordingly.
That would not just limit interoperability and competition. It would fundamentally constrain how broadly AI can be trusted, and therefore how it can innovate, scale and create economic value.
Fadi Chehadé is the former president and CEO of ICANN, the nonprofit that coordinates the internet’s domain name system, and the founder and managing partner of Ethos Capital. He founded RosettaNet, a B2B standards organization, and previously served as senior adviser to the executive chairman of the World Economic Forum.