1Claw Builds the Security Layer for the Day AI Agents Run Wallets and Sign Trades Autonomously

Signup Write New Story 1Claw Builds the Security Layer for the Day AI Agents Run Wallets and Sign Trades Autonomously by Ishan Pandey by Ishan Pandey | @ ishanpandey AI, Web3, Cybersecurity, Startup Funding & Enterprise SaaS. Top Journalist, Analyst & Thought Leadership Media Leader.

AI, Web3, Cybersecurity, Startup Funding & Enterprise SaaS. Top Journalist, Analyst & Thought Leadership Media Leader.

An AI agent that deploys code, pays an invoice or signs a blockchain transaction usually does it with whatever a human would have used, a private key, an API secret or a cloud credential, sitting in its environment where anything that talks to the model can reach it. One prompt injection or one leaked log hands an attacker everything the agent was allowed to do, at a time when CrowdStrike has watched intruders move from first access to the rest of a network in 27 seconds.

1Claw sells a different arrangement in which the agent gets permission to act without receiving the master key, through HSM-backed vaults, per-agent spending limits and an enclave proxy that screens model traffic for prompt injection. It signs transactions across more than 109 EVM chains as well as Bitcoin, Solana, XRP, Cardano and Tron, offers a free tier for developers and has just announced integrations with Bankr, Google Cloud, Circle, Tenderly, Coder and the XRP Ledger.

Today we sit down with Kevin Jones, its Founder and CEO and a veteran of Web3 infrastructure, on what security looks like for software that no longer waits for a human to press the button.

Ishan Pandey: Hi Kevin, welcome to our "Behind the Startup" series. You went from photography to Web2 infrastructure, then to blockchain and now 1Claw. Tell us about your journey and what problems did you see in the market that made you build 1Claw?

Kevin Jones: Throughout my 25-year career in technology, one thing has always motivated me: figuring out how things work, finding ways to make them better, and solving problems that matter. I've also learned that as technology becomes more powerful, keeping it secure becomes increasingly important.

I've worked across companies including Best Buy, Yellow Pages, and NGINX, progressing from Site Reliability Engineer to Solutions Architect. Along the way, I've seen firsthand how complex and fragile technology systems can be, and how a single vulnerability can create serious consequences. Those experiences taught me that security shouldn't be an afterthought. It needs to be part of how we build technology from the beginning.

My curiosity about cryptography eventually led me into Web3. I was fascinated by how cryptography, data ownership, and decentralized governance could change the way we establish trust online. I began to see that some of these ideas could also help solve security challenges in traditional Web2 systems. For me, it was never just about blockchain; it was about finding better ways to protect the systems and people who depend on technology.

Then AI agents started taking off, and I saw another important challenge emerging. We're moving beyond AI that simply answers questions to AI that can access tools, interact with systems, and take action on our behalf. But these agents often need access to sensitive credentials and APIs to do their jobs. If that access isn't properly controlled, the consequences can be serious. I saw an opportunity to apply what I'd learned over the years to help address a problem that will only become more important as AI evolves.

That's what inspired me to create 1Claw. I want to help make AI agents safer to use by giving organizations better control over what they can access and do, while still allowing them to be useful and autonomous. I believe we shouldn't have to choose between innovation and security. We should be able to embrace what AI makes possible while building the safeguards that let people trust it.

Ultimately, what drives me is the opportunity to solve meaningful problems and help make technology work better for everyone. AI has enormous potential, and I want to help ensure that potential can be realized safely.

Outside of technology, photography has always been my creative outlet. It reminds me to slow down, look at things differently, and appreciate the beauty in the world around us. I think that same curiosity carries over into my work: whether I'm behind a camera or solving a complex technical problem, I'm always looking for a new perspective.

Ishan Pandey: You mention that the agent gets permission to act but not the master key. Could you walk us through what actually happens, step by step, when an agent asks 1Claw to sign a transaction or call a paid API on its behalf?

Kevin Jones: The idea is simple: an AI agent should be able to get work done without ever having unrestricted access to the credentials that make it possible.

Let's say an agent needs to call a paid API or sign a transaction. Instead of handing the agent a permanent API key or private key, 1Claw acts as a controlled security layer between the agent and the resource it needs.

First, the agent requests an action. It might need to call an API, access a service, or initiate a blockchain transaction.

Second, 1Claw verifies the agent's identity and checks its permissions. Is this agent authorized to perform this action? Does it have access to this particular service? Are there restrictions on what it can do?

Third, 1Claw applies the relevant security policies. Depending on the configuration, that could mean limiting which credentials the agent can use, which actions it can perform, or what resources it can access.

Fourth, if the request is authorized, 1Claw facilitates access to the required credential or signing capability without giving the agent unrestricted control of the underlying secret. For a paid API, that could mean enabling an authorized API call without exposing the long-lived credential. For a blockchain transaction, it could mean routing a signing request through a protected signing mechanism, subject to the applicable permissions.

Finally, the action proceeds within those boundaries, and the organization can maintain visibility into how access is being used.

The important distinction is that the agent gets permission to perform a specific action, not a master key that gives it unlimited power.

That's the philosophy behind 1Claw: give agents the access they need to be useful, while keeping control of sensitive credentials and permissions outside the agent itself.

Ishan Pandey: CrowdStrike says AI-enabled adversaries increased their operations by 89% last year and the fastest breakout it recorded took 27 seconds. When attackers move at machine speed, which part of a company's security stack fails first? And how do companies build new cybersecurity rails?

The first thing that breaks is often the assumption that humans can keep up. Traditional security relies heavily on people detecting unusual activity, investigating alerts, and responding to threats. But when attackers operate at machine speed, manual processes and fragmented security systems can struggle to keep pace.

AI agents introduce another challenge: they don't just process information; they can take action across multiple systems using credentials and permissions. If those credentials are compromised or an agent is manipulated, the potential impact can spread quickly.

Companies need a new security layer built for autonomous systems. That means giving every agent a verifiable identity, enforcing least-privilege access, protecting credentials, monitoring actions, and being able to revoke access quickly when something goes wrong.

At 1Claw, we're focused on this access-control problem: helping organizations manage what AI agents can access and do without exposing sensitive credentials or giving them unnecessary permissions.

When threats move at machine speed, security can't depend on humans reacting fast enough. It has to be built into the infrastructure from the start.

Ishan Pandey: Bankr and Circle put agents next to real money, Tenderly simulates transactions before they are signed and Coder hands coding agents a cloud workspace. What can an agent do safely through these integrations today that it could not do a year ago?

Kevin Jones: The big shift is that AI agents are moving from making recommendations to taking real actions. With platforms like Bankr and Circle, agents can interact with wallets and move money. Tenderly lets developers simulate transactions before signing them, while Coder gives coding agents controlled environments to write and execute code.

That means agents can now participate in real financial and development workflows with safeguards that weren't as accessible a year ago.

But enabling an agent to act is only half the problem. How do you make sure it can do only what it's supposed to do? Simulating a transaction or isolating a workspace helps, but neither automatically prevents an agent from accessing the wrong credentials or exercising excessive permissions.

That's where 1Claw comes in. We're focused on helping organizations control which credentials, APIs, and tools their agents can access, limit their permissions, and revoke access when needed.

The future isn't about giving agents unlimited autonomy. It's about giving them the authority to get work done without giving them the keys to everything.

Ishan Pandey: The FBI attributed the $1.5 billion Bybit theft to North Korea after attackers fooled the people approving a transaction rather than breaking the keys. If the weakest point is the moment of approval, how do you protect a decision that an agent makes on its own?

The $1.5 billion Bybit theft exposed a critical weakness in security: attackers can exploit the process used to approve a transaction, even when the underlying systems have strong protections. The FBI attributed the attack to North Korea, highlighting how manipulating a trusted workflow can have devastating consequences.

With AI agents, the challenge is even greater: how do you secure a decision when the agent makes it autonomously? You can't rely on the agent to recognize every malicious instruction or judge whether its own actions are safe.

The answer is to separate the agent's ability to request an action from the authority to execute it. Every request should be checked against independent security policies: What is the agent trying to do? Does it have permission? Is the action within its limits? If not, it should be blocked. High-risk actions can also require human approval.

That's what we're focused on at 1Claw: giving agents the access they need to get work done without giving them unrestricted authority over credentials, APIs, and critical systems.

The principle is simple: let agents make decisions, but make sure security controls, not the agents themselves, determine what they're allowed to do.

Ishan Pandey: Imagine an agent starts misbehaving at three in the morning while nobody on the team is watching. How fast can a company revoke it through 1Claw and who is accountable for what it has already done?

Kevin Jones: If an AI agent starts misbehaving at three in the morning, the company shouldn't have to wait for someone to wake up to contain it. Access needs to be revocable as soon as a problem is detected, cutting off the agent's ability to use protected credentials and resources. The exact response time depends on how the system is configured and where the agent is operating.

With 1Claw, our focus is giving organizations control over what agents can access and do, including the ability to revoke their permissions when necessary. Just as importantly, revocation doesn't erase what an agent has already done. That's why visibility and auditability matter: organizations need to understand which actions were taken, what resources were accessed, and which permissions allowed them.

Revoking access is a technical control; accountability is an organizational responsibility. The company deploying the agent remains responsible for its permissions, policies, and oversight. 1Claw's role is to help make access controllable and actions traceable, so organizations can contain incidents and understand what happened.

Autonomous agents may operate around the clock, but security and accountability can't depend on someone being awake to supervise every action.

Ishan Pandey: Last question. Which lesson about keys did crypto learn the hard way that the rest of the AI industry is about to learn all over again?

Kevin Jones: The biggest lesson crypto learned the hard way is that whoever controls the keys controls everything, and if those keys are compromised, the damage can be irreversible. In Web3, we saw billions of dollars lost through stolen private keys, exposed credentials, compromised wallets, and poorly managed access. The technology could be incredibly sophisticated, but security often came down to a simple question: who holds the keys, and what can they do with them?

Now we're seeing a similar challenge emerge with AI agents. We're giving agents access to APIs, credentials, databases, and business systems so they can act autonomously. But if an agent has unrestricted access, gets manipulated, or exposes its credentials, the consequences can extend far beyond a single task.

The lesson crypto learned is that securing the technology isn't enough. You have to secure access to the technology.

That's one of the ideas behind 1Claw. AI agents need credentials to be useful, but they shouldn't have unlimited access to everything those credentials unlock. We need to control what agents can access, limit their permissions, protect their secrets, and be able to revoke access when something goes wrong.

Crypto learned that key management isn't a secondary security feature; it's fundamental infrastructure. The AI industry has an opportunity to learn that lesson before compromised agents turn it into another expensive one.

Don’t forget to like and share the story!

AEREDIUM Introduces AI Guardrails to Keep Autonomous Agents in Check When Handling Payments and Code