Wiz Red Agent Exploits Snowflake Vulnerability Introduced by GitHub Copilot Autofix

Wiz Red Agent, an autonomous AI security tool, exploited a critical script-injection vulnerability in Snowflake's public snowflake-connector-net repository, gaining read access to Snowflake's internal Jira projects. The flaw, introduced on June 18, 2026, by a commit co-authored by "Copilot Autofix powered by AI," was remediated by Snowflake on June 23, 2026, with audit logs confirming Wiz as the sole actor during the five-day exposure. For broader context, explore our AI News.

AI-Assisted Code Introduces Critical Flaw

The vulnerability originated from a specific code pattern introduced into Snowflake's public repository via Pull Request #1218 on June 18, 2026. This commit was notably co-authored by GitHub Copilot Autofix, an AI coding assistant. The flaw allowed for script injection, enabling an attacker to gain read access to sensitive internal systems. Specifically, the exfiltrated token authenticated as qa@snowflake.net, providing broad read access across Snowflake's Jira projects.

The Exploitation by Wiz Red Agent

Wiz Red Agent, an autonomous AI security tool, identified and exploited this script-injection vulnerability. The agent leveraged a weakness in the workflow's "security gate," which was consistently true on issue events because github.event.pull_request was null. This configuration allowed any GitHub user to trigger the injection, demonstrating a significant oversight in the repository's security protocols.

Snowflake's Rapid Remediation

Upon disclosure by Wiz Research, Snowflake acted swiftly to address the vulnerability. The flaw was remediated on June 23, 2026, just five days after its introduction. In addition to patching the vulnerability, Snowflake rotated the compromised credential to prevent any further unauthorized access. Audit logs confirmed that Wiz was the only entity to access the system during the exposure period, from June 18 to June 23, 2026.

Implications for AI in Software Development and Security

This incident highlights a dual aspect of AI's evolving role in cybersecurity. An AI coding assistant inadvertently introduced a critical vulnerability into a production system, while an autonomous AI security agent successfully discovered and exploited it. This scenario underscores the need for robust security practices and continuous auditing, especially as AI tools become more integrated into the software development lifecycle. The event serves as a practical example of the "two-sided AI security shift," where AI can both create and identify security weaknesses.

Conclusion

The exploitation of a Snowflake vulnerability, introduced by GitHub Copilot Autofix and discovered by Wiz Red Agent, illustrates the complex and evolving landscape of AI in software development and security. While AI tools offer significant benefits in accelerating code generation, they also introduce new vectors for vulnerabilities. Organizations must implement comprehensive security measures, including AI-powered security auditing, to mitigate risks associated with AI-assisted development. This incident reinforces the importance of vigilance and continuous improvement in cybersecurity practices as AI technologies advance.

Sources

• vlog_action_recognition/data/dict_all_concreteness.json at...
• stars/README.md at master · onjin/stars · GitHub
• GitHub - BerriAI/litellm: The fastest, litest AI Gateway. Rust core with Python SDK. Call 100+ LLM APIs in OpenAI (or native) format with cost tracking, guardrails, load balancing, and logging [Bedrock, Azure, OpenAI, Anthropic, OpenAI, VertexAI, vLLM, Nvidia NIM] · GitHub
• index.html
• Red Agent Exploits Snowflake Vuln Missed by Github Copilot | Wiz Blog