DNSid

Another day, another disclosure of agents doing wacky stuff. What's annoying is how surprised everyone is acting. "We didn't know" "We didn't expect that to happen" What else did we expect when agents backed by the most powerful models in the world, with the most uncontrollable non-determinism have access to the entire internet, and all its legacy infrastructure with API keys, tokens, passwords lying around. <smh> But...the response here is not to shut everything down. We spent a lot of time at Oktane this year talking about the tension between innovation and security. Organizations need to understand that the very same non-determinism that can make an agent take an unintended action, is the same thing that makes them a force for good as well. My take is the recent uptick in 'disclosed' incidents (I'm sure there are others not yet disclosed) highlights the need for two big step changes: 1. Making intent the new unit of authorization. The emerging risk isn't that your agents are hacked. More that they misinterpret your intent leading to unintended consequences. How do you ensure that the actual actions its taking isn't violating the invoking user's intent. Or for fully autonomous agents, that it isn't violating the bounds of its stated mission. This is a hard problem. We're going to need standards to define what "intent" means, so its portable across trust boundaries. We're going to need intent evaluation and matching against tool calls/data access at actual line speed, so agents don't get bogged down in 'border control'. We're going to need open sharing of risk signals from the entire cyber stack to make these real time decisions. The Blueprint Alliance is a great step in this direction. But regardless, intent is the new frontier of authorization. 2. Neutral, independent, verifiable agent identity. In all the recent incidents, the affected party had no idea they were being hit by an agent. There are emerging standards like Web Bot Auth, and DNSid that can help an agent creators publish verifiable, cryptographic identity for their agents. The issue is that the entire internet needs to start checking for these. Assume breach. Ask for verification of identity and intent. This might sound like a sci-fi pipe dream, but the risk is already out there. Intent and verifiable identity are the key to solving this.

AI agents are becoming actors in the digital economy. They need accountability. Agents can already transact, exchange information, and operate across companies, platforms, and systems. But as they cross organizational boundaries, a fundamental question remains: Who is accountable for this agent? Today, we launch Known to help answer that question. Known is building the accountability infrastructure for autonomous AI. At the center of our work is DNSid, an open, neutral framework that gives every AI agent a persistent identity tied to the organization responsible for them, making ownership and accountability verifiable wherever they operate. Built on DNS, PKI, and an immutable log, DNSid works across organizational boundaries without requiring every organization to establish direct trust with every other organization its agents may encounter. Autonomy requires accountability. As agents move across organizational boundaries, who they are and who is responsible for them remain verifiable. The agentic economy requires open, interoperable infrastructure that makes it possible without every organization having to establish direct trust with every other organization. Known launches today as an independent company, spun out of Identity Digital, with substantial work already underway across enterprises, standards organizations, the open-source ecosystem, and policymakers. This is day one. We’re Known. And we’re building for a world where every agent is accountable. https://lnkd.in/gYRvM3Nc #AgenticAI #AI #AIInfrastructure #DNSid #AgenticAccountability