ArgusRed

ArgusRed is an automated security scanning product built by Cosine, a model lab backed by Y Combinator and UK Sovereign AI. It connects to a GitHub repository and attacks the code the way a real intruder would, reporting only exploits it could actually reproduce — not a list of unverified flags. The service runs on UK and EU servers and is positioned for small teams that need pen-test-grade evidence without a dedicated security team.

What It Is

ArgusRed is an AI-driven penetration testing service that sits in the automated security testing category. Unlike traditional static analysis scanners that surface hundreds of potential vulnerabilities with confidence scores, ArgusRed's agent hunts for exploitable paths, then stands up an isolated sandbox mirroring the target stack and attempts to reproduce each candidate. Only confirmed exploits — with the actual request and response — reach the developer. The product is built on a security-tuned model that Cosine post-trained specifically to reason like an adversary, rather than wrapping an off-the-shelf API.

How the Workflow Operates

The process runs in three steps:

• Connect your repo — One-click GitHub authorization scoped to a single chosen repository; no configuration or procurement required.
• Attack phase — The agent searches for exploitable paths and reproduces each candidate in a sandbox that mirrors the repository's dependencies and stack.
• Receipt delivery — A pen-test-grade report is produced containing the confirmed exploit, the exact request sent, the response received, and a pull request that patches the vulnerability and passes existing tests. Cosine states the service re-attacks after the fix is merged to confirm closure.

Architecture and Provenance

ArgusRed runs on a model Cosine post-trained for adversarial security reasoning — not a prompt wrapper around a general-purpose API. The homepage explicitly distinguishes this as a differentiator in what it calls a "is this a wrapper?" market. All source code and sandbox execution stay on UK and EU infrastructure, and a plain-language data-handling statement ships with every report.

Audience and Use Case

The product is aimed at small engineering teams — particularly those with enterprise customers asking security questions — who need verifiable proof of exploitability rather than a triage backlog. The homepage frames the core problem as scanner fatigue: teams that have already muted high-volume scanners because the signal-to-noise ratio is too low. ArgusRed's answer is to discard everything it cannot prove before it reaches the developer.

Backing and Provenance Signal

Cosine, the lab behind ArgusRed, is listed as a Y Combinator company and appears in UK Sovereign AI portfolio materials and GOV.UK announcements. An NVIDIA blog post on UK Sovereign AI advancements is also cited on the homepage. ArgusRed is described as built on the same engine as the Cosine coding agent.