Fadi Chehadé, the former CEO of ICANN who oversaw the internet’s domain name system from 2012 to 2016, published an opinion piece in Newsweek on July 1 arguing that AI agents need open accountability standards before proprietary solutions fragment the ecosystem beyond repair.
The Core Argument
Chehadé frames the problem around three questions that lack shared answers: what agents can do (capability scoping and permission boundaries), where they can operate (system boundaries and network access), and who is responsible when they act (liability chains and principal-agent responsibility). Without industry-wide answers, he argues, every platform and enterprise will invent its own local controls, solving their immediate problems while making agents harder to trust across companies, markets, and borders.
“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,” Chehadé wrote.
He specifically cited OpenClaw as evidence that controlling agents “once they can operate across applications with real permissions” remains difficult, and pointed to Anthropic’s Claude Mythos model and Project Glasswing as examples of frontier models becoming powerful enough to identify vulnerabilities in critical software.
The DNS Parallel
Chehadé draws directly on his ICANN experience. Before DNS created a common naming foundation, the internet was a collection of isolated networks where interoperability depended on prior agreements rather than shared standards. DNS provided a neutral layer that enabled independent systems to work together without a central authority controlling them. According to Wikipedia, Chehadé led ICANN during its TLD registry expansion and oversaw the initiation of the IANA functions transfer from US government control.
His argument: AI agents are at an equivalent moment. An agent may have one identity inside a cloud platform, another inside an enterprise environment, and different credentials across each tool it touches. No durable ownership record persists across the ecosystem. Without a shared accountability layer, trust fragments by default.
The Timing Context
The piece lands during a week of parallel standards activity. On June 29, a coalition including Google, IBM, Circle, UiPath, Stellar, Cardano, and Hedera launched the Legal Context Protocol (LCP), an open-source Apache 2.0 framework for AI agents to transact on behalf of humans with verifiable legal terms. The same week, China’s State Administration for Market Regulation released a national standard for AI agent interconnection with a unified identity management framework. Senator Mark Warner introduced the AI Agent Act, the first federal bill requiring an FTC registry of trusted AI agents.
These are not coordinated efforts. They are parallel responses to the same pressure Chehadé describes: agents operating across organizational boundaries need shared trust infrastructure, and the window to establish open standards is narrowing as proprietary solutions harden.
What Chehadé Gets Right, and What He Omits
The diagnosis is precise. Chehadé’s background in internet standards gives him credibility on the pattern: open infrastructure enabled competition and innovation; closed alternatives would have fragmented the market. The same structural risk applies to AI agents.
What the piece does not address is who builds and governs these standards. DNS succeeded partly because ICANN existed as a multi-stakeholder body with government, industry, and civil society participation. No equivalent body exists for AI agent accountability. The LCP coalition is industry-led. China’s SAMR standard is government-led. Warner’s AI Agent Act is legislative. Each approach embeds different assumptions about who controls the rules.
The question Chehadé poses is the right one. The answer will depend on whether any single approach gains enough adoption to become the default before the window closes.