Vint Cerf aims to assign an identity to AI agents.

      Soon, the internet will be populated by AI agents acting on our behalf. Currently, there is no dependable method to identify who is behind these agents. Vint Cerf, one of the pioneers of the internet, aims to address this issue.

      Cerf co-developed TCP/IP, the protocol that allows different systems on the internet to communicate. He recently left Google after two decades and is now joining the advisory council of Innovation Labs, a group that is creating an open identity layer for AI agents, as announced by the company.

      The missing component

      The issue is straightforward. Most AI agents currently operate within a single company's systems. However, companies want these agents to navigate the open web and interact directly with one another. There is no common method to verify ownership of an agent or accountability for its actions.

      Innovation Labs is a branch of Identity Digital, which manages domain-name registries. Their proposal, known as DNSid, aims to provide each agent with a permanent identity linked to an existing domain name, supported by cryptographic validation. They have already submitted the design to the primary standards organization for the internet.

      Reason for Cerf's involvement

      Cerf views this as the next significant architectural challenge for the internet. He mentioned to TechCrunch, “The question revolves around what authorities they possess, the origins of those authorities, and who is responsible.”

      He anticipates a complex situation. “It’s going to be a captivating and potentially frustrating time,” he remarked. Competing standards are already emerging, and Cerf believes that none will succeed based solely on politics, but rather on what is effective, similar to the progression with TCP/IP.

      Maintaining openness

      The argument is that no single tech giant should control the standard. Innovation Labs asserts that it will not handle the registration data directly. “There’s significant resistance to a hyperscaler releasing a standard and possessing that proprietary data,” interim leader Allie Kline shared with TechCrunch. The organization claims it is currently testing the system with various unnamed cloud providers.

      An agent-driven internet

      The urgency increases as agents are rapidly proliferating, from Amazon's updated Alexa to various enterprise applications, and they have already begun to cause issues. Researchers have managed to manipulate them into exposing private code and executing substantial ransomware attacks. Regulatory bodies are also responding, with China's new agent rules and Delaware's initiative to grant agents a legal identity.

      Cerf is uncertain if an internet run by agents is inevitable. However, he believes efforts will be made to construct one nonetheless. “We are inherently lazy beings,” he remarked. If an agent can perform a task for us, we are likely to allow it.