Wiz Says Its Red Agent Exploited a Snowflake Vulnerability That GitHub Copilot Autofix Missed

Cloud security provider Wiz has published research claiming that its AI agent, Red Agent, autonomously discovered and exploited a GitHub Actions vulnerability in one of Snowflake's public repositories, a flaw that GitHub Copilot Autofix had missed when reviewing the code.

According to Forbes, Wiz says GitHub Copilot Autofix acted as a co-author that approved the code change without noticing the critical vulnerability, though it is unclear whether the change itself was AI-assisted. The vulnerability was disclosed through Snowflake's HackerOne platform and mitigated on June 23, but before the fix, Wiz used it to reach sensitive data inside Snowflake's internal Jira environment.

Concretely, Red Agent abused a script injection flaw in snowflakedb/snowflake-connector-net that lets an unauthenticated user execute arbitrary commands inside a GitHub Actions runner by opening a GitHub issue with a specially crafted title.

The incident cuts both ways: AI models can now surface and exploit vulnerabilities without human intervention, while AI coding assistants can simultaneously miss critical flaws in the CI/CD pipeline.

Wiz Red Agent went generally available in July and is now used by about 40% of Wiz customers, scanning millions of assets every month. Gal Nagli, head of Offensive Security at Wiz, said the flaw was autonomously found and exploited by Wiz's AI without human intervention, showing that frontier models can already exploit supply chain risks by themselves.

Nagli also warned that defenders cannot trust AI code generation to be fully autonomous, or they face additional risk, urging companies to proactively scan with AI because frontier models can execute like autonomous experts end to end. Erik Avakian, a technical counsellor at Info-Tech Research Group, framed cybersecurity as an emerging AI-versus-AI battlefield, where AI-assisted development expands the attack surface even as it speeds up delivery.

The research arrives less than a month after Wiz announced Project Atlas, a vulnerability scanning solution that uses multiple AI models. The next questions are how enterprises fold AI-driven proactive scanning into development workflows, and how governance and regulation adapt to a world where autonomous attacks are becoming routine.