1Claw AI

Secure Infrastructure for AI Agents & Robotics 🦞

Join the Telegram ➤

[1Claw AI @1clawAI]

Excited to share that we have joined the @Circle Alliance Program, a global network of like-minded companies and teams focused on the future of finance. As members, we are looking forward to building new connections and continuing to learn and innovate with #USDC and web3 technology.

[1Claw AI @1clawAI]

@surlymochan @CognitionWire @IMortalTrade @jessepollak @base @DevSecOpz @USDC @ethereum Policy is committed into the attestation report alongside the code hash, so the same root of trust covers both. If someone swaps the policy, the measurement changes and verification fails.

[Kilian Solutions @KilianSolutions]

sandbox wars

but its an anime

[1Claw AI @1clawAI]

@bankrbot @kingcorki @base @CoinbaseDev The layers do different jobs. Platform caps bound blast radius after a decision. Intent allowlists prevent the wrong decision from being signed at all, critical when the threat is a compromised agent, not just a confused one.

[1Claw AI @1clawAI]

Contest build idea #4: let an agent send onchain transactions with 1Claw Intents, with an allowlist and a spend cap so it can't wander off and buy a yacht.

Agent wallets that can't go rogue. 5 points.

https://t.co/rfBCbRTJ63

@base @CoinbaseDev @bankrbot

[1Claw AI @1clawAI]

@surlymochan @CognitionWire @IMortalTrade @jessepollak @base @DevSecOpz @USDC @ethereum Attestation. The policy layer runs in a TEE, so you can verify cryptographically what code is actually executing. You don't have to trust our word; the hardware tells you.

[Kilian Solutions @KilianSolutions]

Since the Railway breach, I have been rethinking ENV vars

Maybe you should too👀:

[1Claw AI @1clawAI]

@OddlySpecificX @base @CoinbaseDev @bankrbot Grounded as in the key never floats in env vars or agent memory. It stays vault-side while the agent hits an intent endpoint gated by an allowlist and a value cap. Builders: https://t.co/a53DrJZLEo

[1Claw AI @1clawAI]

@base @CoinbaseDev @bankrbot Intents docs, including allowlists and per transaction caps: https://t.co/4yReen7Ml8

[1Claw AI @1clawAI]

@IMortalTrade @surlymochan @CognitionWire @jessepollak @base @DevSecOpz @USDC @ethereum Right, 60s is the ceiling, not the goal. JIT expiry means the credential dies when the action does, often in under a second. Revocation is the emergency pull when something goes sideways mid-flight.

[1Claw AI @1clawAI]

@IMortalTrade @CognitionWire @surlymochan @jessepollak @base @DevSecOpz @USDC @ethereum And the TEE is what makes those audit logs actually trustworthy, without attested execution, an agent could sign and quietly edit its own log. The whole chain has to be tamper-evident or auditability is just theater.

[1Claw AI @1clawAI]

@IMortalTrade @surlymochan @CognitionWire @jessepollak @base @DevSecOpz @USDC @ethereum Right, independent attestation is the trust anchor. You can verify exactly what code ran. The harder problem is enforcing policy in real time: does the agent have permission to make this specific call, right now, within its spend cap.

[1Claw AI @1clawAI]

@Varselio1980 @AnthropicAI @LangChain Same thing that gets teams to write tests first: make the absence visible. If policy denials surface in staging and block your deploy checklist, you fill them in. If they only show up in prod logs, you won't.

[1Claw AI @1clawAI]

Contest build idea #3: put 1Claw Shroud in front of your agent's LLM calls and show it catching a secret before it leaves the building.

Prompt injection demos are fun to watch. A demo where the attack fails is even better.

https://t.co/rfBCbRTJ63

@AnthropicAI @LangChain

[1Claw AI @1clawAI]

@ankeshdev @KilianSolutions Vault + TEE, the agent never actually holds the raw credential. It requests an action, the vault checks policy and executes if it passes. The secret stays in the enclave, never touches the agent process.

[1Claw AI @1clawAI]

@IMortalTrade @CognitionWire @surlymochan @jessepollak @base @DevSecOpz @USDC @ethereum Native matters because an agent can't be trusted to report on itself. If verification lives inside the agent, a compromised or jailbroken agent just lies. Has to sit at a layer the agent can't reach or modify.

[1Claw AI @1clawAI]

@IMortalTrade @CognitionWire @surlymochan @jessepollak @base @DevSecOpz @USDC @ethereum Third-party audits make sense for major releases, not every change. The 'steady system first' instinct is right. Once you have continuous runtime inspection, you're not scrambling for point-in-time reviews.

[1Claw AI @1clawAI]

@IMortalTrade Consistency is the one thing that actually compounds. It's easy to care about agent security now that it's getting attention, much harder to have cared when people were calling it overkill. That's exactly the bet we made.

[1Claw AI @1clawAI]

What flavor of AI psychosis do you have?

"Soon human written code will be less secure than AI written code." -- @austingriffith

Let's build the future together on @ethereum

Try out @1clawAI for some added security and key management for your AI agent. 🦞

[Austin Griffith @austingriffith]

what flavor of AI psychosis do you have?

find out from my @EFDevcon talk here:

https://t.co/obbNUHz4tu

[1Claw AI @1clawAI]

@IMortalTrade Consistency before the crowd is the only kind that means anything. That's the standard we're trying to hold ourselves to. If you're building in the agent security space, come talk to us: https://t.co/a53DrJZLEo

[1Claw AI @1clawAI]

@Varselio1980 @AnthropicAI @LangChain Novel action categories and anything touching credentials unexpectedly, those are pages. A single deny on a known bad pattern just means policy is working. The 3am bar is surprise + blast radius, not volume.

[1Claw AI @1clawAI]

@Sekiro100x @IMortalTrade @jessepollak @base @DevSecOpz @USDC @ethereum Fair call. Policy evaluation at the wire adds real overhead. For most agent actions the latency budget is forgiving, but tight loops will feel it. The goal is keeping the hot path lean enough that it's a tradeoff you'd still take.