Visa ships a security AI that patches production code before any human reviews it
Visa’s open‑source Visa Vulnerability Agentic Harness (VVAH) now runs an 11‑stage autonomous loop that scans code, writes a fix, and validates the patch before any human review, shipping as the default configuration. The harness, built on Anthropic’s Claude Mythos model, has grown from 595 ★/97 forks on July 20 to over 2,300 ★/300 forks by August 25, with a reported clone‑to‑visitor ratio of roughly 9 %. Engineers can integrate VVAH to automate discovery, remediation, and validation of vulnerabilities, shifting the bottleneck from detection to fixing and proof‑of‑concept validation.
⚡ Key Takeaways
- The VVAH pipeline executes **11 distinct stages** and edits source files automatically unless capped at detection.
- It leverages **Anthropic’s Claude Mythos** model for semantic reasoning and exploit generation.
- The default autonomous operation raises a **trade‑off**: faster vulnerability remediation versus loss of a human authorization gate.
- Integration is via the **open‑source VVAH repository on GitHub**, which includes the abstract syntax tree call‑graph scanner.
- The repo **does not accept external code contributions**, meaning the harness cannot ingest third‑party patches.
🔧 Tools & Libraries
Automating the full fix‑validate cycle lets security teams close vulnerabilities orders of magnitude faster, but teams must weigh the risk of unsupervised code changes against the speed gains.
✅ Practical Steps
- Clone the VVAH repository from GitHub and run the default pipeline to let it scan, fix, and validate code automatically.
Want the full story? Read the original article.
Read on VentureBeat AI ↗