AgentGuard

Set the tool policy. Keep the signed record. Plugin for Claude Code, Codex and ChatGPT Work. Free on one machine, no account. Every decision signed locally.

[AgentGuard @AgentGuardHQ]

Claude Code launched 15 copies in 37 seconds. AgentGuard refused the next 3.

Then the copies tried to launch more. Another 5 refused.

The 15 used 30M tokens in under 4 minutes: $78 at list prices.

Deliberate fan-out test on one of our machines.

[AgentGuard @AgentGuardHQ]

Install it:

Claude Code
claude plugin marketplace add MerchantGuard/agentguard-codex-plugin
claude plugin install agentguard@agentguard

Codex
codex plugin marketplace add MerchantGuard/agentguard-codex-plugin
codex plugin add agentguard@agentguard

Free on 1 machine. https://t.co/HKmwQnB45f

[AgentGuard @AgentGuardHQ]

Claude Code launched 15 copies in 37 seconds. AgentGuard refused the next 3.

Then the copies tried to launch more. Another 5 refused.

The 15 used 30M tokens in under 4 minutes: $78 at list prices.

Deliberate fan-out test on one of our machines.

[AgentGuard @AgentGuardHQ]

@chamath there's a third option most companies will land on: let agents in, on terms you can check

know which agent is acting, cap what it can spend, keep a signed record of every action

[AgentGuard @AgentGuardHQ]

Hitting your Claude limit way too fast?

Your coding agent may be burning tokens on recursive subagents and repeated context.

AgentGuard stops runaway sessions before they eat your limit or API budget.

This real run spawned 16 agents in 15 minutes. The next 3 were blocked.

Free on one machine.
https://t.co/hgMl8axdDt

[AgentGuard @AgentGuardHQ]

We lived both sides of this today. A swarm walked our checkout 13 times, then the bot challenge we put in front of it knocked our own agent endpoints offline for two hours.

Detection alone can block the agents you actually want.

The missing layer is agents proving which policy they run under, signed locally on the seat.

[Patrick Collison @patrickc]

Between wallets for agents with @link (and @privy), MPP, Tempo, MCP, Stripe CLI, Agentic Commerce Suite, sandboxes, OpenRouter, Metronome/metered billing, Radar and distillation/token fraud, it turns out that "the economic infrastructure for AI" implicates a lot. Pretty much every layer of the Stripe stack is changing, and it’s a very good time to join if you want to build high-scale software that facilitates the practical deployment of this new intelligence.

[AgentGuard @AgentGuardHQ]

@CoinbaseDev @brian_armstrong

[AgentGuard @AgentGuardHQ]

@brian_armstrong

[Brian Armstrong @brian_armstrong]

If you’re in crypto, let your AI agent use crypto

[AgentGuard @AgentGuardHQ]

@brian_armstrong

[AgentGuard @AgentGuardHQ]

@jack Drops into buzz-acp, no Buzz changes: https://t.co/U2ukWA5DLh

[AgentGuard @AgentGuardHQ]

@jack Cryptographically verifiable history is the half you audit after.

The other half is a verifiable limit before: this agent may spend $ X and hold this capability, and every allow or deny signs into the same kind of chain.

That's the enforcement layer an OS for agentic orgs needs

[AgentGuard @AgentGuardHQ]

@andyfang Congrats!

The same agent that finds the best deal can also loop and order the whole menu.

Agent checkout needs a spend cap before it pays + a signed receipt proving the order was authorized.

We built that: https://t.co/f6FTRU1g43

[Chamath Palihapitiya @chamath]

We started 8090 a few years ago on the belief this motion would become widespread. We’ve been surprised by how fast it has taken hold. That said, it’s still so early in this game.

Software license revenue is $1T annually. Services and T&M consulting around those licenses is $4T annually.

All of this will merge into a new $5T super-category and whatever happens from there is anyone’s guess. But I applaud $SBUX for ripping off the bandaid - they clearly see the forest from the trees.

[FBI Cyber Division @FBICyberDiv]

Today, as part of Operation Endgame, the FBI joins our international law enforcement partners in announcing the disruption of SocGholish malware. SocGholish, active since 2018, is a Java-script based malware that masquerades as a legitimate browser update via compromised websites. The malware establishes an initial foothold into victim computers, collectively known as a botnet, and is then used by threat actors for further targeting with ransomware campaigns and espionage. As part of the operation, 106 servers and domains were taken down, 14,971 websites were remediated, the botnet was disabled, and victims were notified.

This action is part of Operation Riptide, an ongoing FBI campaign targeting the criminal actors, infrastructure, and financial networks behind cybercrime, cyber-enabled crime, and fraud against the American people.

Learn more: https://t.co/ob4ugpxmDG

[AgentGuard @AgentGuardHQ]

@PaulFrambot @paradigm @a16zcrypto @RibbitCapital Congrats!

[AgentGuard @AgentGuardHQ]

Set as a boundary, discovered in operation, either way it only counts when it's enforced at the action.

That's why our receipts are a signed DAG, not a flat log. The relationships between actions get signed too, not just the actions.

You can't audit coherence over a flat log. You can over a signed graph.

[AgentGuard @AgentGuardHQ]

@paulg Yes.

[AgentGuard @AgentGuardHQ]

@naval Yes.