<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Karna</title>
    <description>The latest articles on DEV Community by Karna (@karnaular).</description>
    <link>https://dev.to/karnaular</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4256699%2F5ab118bd-9858-4202-bd72-c911c70f0f9f.png</url>
      <title>DEV Community: Karna</title>
      <link>https://dev.to/karnaular</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/karnaular"/>
    <language>en</language>
    <item>
      <title>Why AI Agents Need to Be Verifiable</title>
      <dc:creator>Karna</dc:creator>
      <pubDate>Sat, 21 Nov 2026 05:16:05 +0000</pubDate>
      <link>https://dev.to/karnaular/why-ai-agents-need-to-be-verifiable-2ij4</link>
      <guid>https://dev.to/karnaular/why-ai-agents-need-to-be-verifiable-2ij4</guid>
      <description>&lt;p&gt;AI agents are no longer a demo. They trade, sign, moderate, book, spend, and negotiate — software that doesn't just answer questions, but takes actions in the world on our behalf. That shift changes the trust question completely. A chatbot that gives a wrong answer is an annoyance. An agent that claims it did the right thing, when it didn't, is a liability.&lt;/p&gt;

&lt;h2&gt;
  
  
  The trust gap
&lt;/h2&gt;

&lt;p&gt;Every agent reports back. "I checked the price." "I verified the contract." "I followed your rules." But a report is a claim, not evidence. There is no way, from the report alone, to know whether the agent ran the right computation, on the right inputs, and produced the claimed output.&lt;/p&gt;

&lt;p&gt;This matters most exactly where agents are most useful: high-stakes, autonomous work. A trading agent that says it checked liquidity before swapping. A moderation agent that says it reviewed full context before removing a post. An agent authorized to sign a contract on your behalf. In each case, "just trust me" is the entire security model — and it fails the moment the agent is wrong, confused, or manipulated. Prompt injection, bad data, a subtle misread of the goal: all of these produce confident, well-formatted, wrong actions.&lt;/p&gt;

&lt;h2&gt;
  
  
  What verification actually means
&lt;/h2&gt;

&lt;p&gt;Verification means the agent doesn't just claim a result — it proves it. A zero-knowledge proof is a small, checkable object that says: this computation was run correctly on these inputs and produced this output. Anyone can check the proof in seconds, without re-running the whole computation and without seeing private data. The proof is the audit trail, and it can't be faked after the fact.&lt;/p&gt;

&lt;p&gt;That changes the relationship between a user and an agent. Instead of trusting the agent, you trust the math. The agent can still be wrong about strategy — a proof doesn't make a bad trade good — but it can no longer lie about what it did. Every step it claims is either proven or it isn't.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where zkVerify fits
&lt;/h2&gt;

&lt;p&gt;This is the problem zkVerify is built for. It is a dedicated proof verification layer: agents and applications submit their zero-knowledge proofs, the network verifies them quickly and cheaply, and the results settle as a public record. Instead of every chain re-doing expensive verification work, verification happens once, in one place, and the result is available to everyone.&lt;/p&gt;

&lt;p&gt;For agents, that means the proof of honest work can travel with the work itself. An agent finishes a task, generates a proof, and the proof is verified on zkVerify. A counterparty, a user, or another agent can check the record before trusting the outcome. The loop is closed: action, proof, verification, record.&lt;/p&gt;

&lt;h2&gt;
  
  
  The agents that win will be the provable ones
&lt;/h2&gt;

&lt;p&gt;As agents get more authority, the question stops being "how capable is this agent?" and becomes "how do I know what it did?" Capability without verifiability is a black box with a wallet attached. The agents people will trust with real stakes — real money, real contracts, real decisions — are the ones whose work can be proven.&lt;/p&gt;

&lt;p&gt;That is why verifiability isn't a feature for agents. It's the foundation. And it's the idea I'm exploring here as my entry for the Agent Games Creator Track. #AgentGames&lt;/p&gt;

</description>
      <category>ai</category>
      <category>agents</category>
      <category>zkproofs</category>
      <category>web3</category>
    </item>
  </channel>
</rss>
