We deployed 10,000 background agents to security-scan our codebase. The system is simple, scales with compute, and runs on publicly available models. From the scan, we fixed several high-severity vulnerabilities.

The scan pipeline is model-agnostic, and does not require a frontier model to drive it. We evaluated several models against our confirmed vulnerabilities, and found that cheaper open-weight models still surface high-severity issues.

We built this to earn trust from Ramp customers, who rely on us for their cards, expenses, and payments. If you have a background coding agent, you can build a similar scan for your customers. Full article:

@beehiiv you’ll love this one

Ramp Labs 🤝 keeping Ramp secure

We have mythos at home?

🐝!!

Project beewing incoming

Don't ask the fox to guard the henhouse. Deploy actual defenses.

I wish I could get 10,000 agents to help me develop my new compute type

whens the ramp labs merch dropping?

Go bees!!

Love to get in contact in case you want persistent, linux microvms that boot in under 30ms and sleep when idle.

@tryramp Vaultfire could hold those agents accountable 😉

Share prompts if you can!

triage is the expensive half of this one. running 10k scanners is cheap, sorting their output isn't

the cheap models finding is the real headline. everyone's obsessed with frontier models for agents but the bitter lesson keeps proving itself. what was the false positive rate like? 10k agents must generate a ton of noise

Kudos! Many will be surprised, as cyber agents like these (Blue Agent Swarms) will surpass coding agents as the dominant Enterprise AI workload.

a new category of defensive infrastructure

10,000 background agents is the interesting bit: security scanning is one of the few agent use cases where parallelism can turn directly into more coverage.