Vulnerabilities in AI Coding Tools Expose GitHub Secrets to Attackers

Scworld

Vulnerabilities in AI Coding Tools Expose GitHub Secrets to Attackers

First seen 8 Aug 2026, 01:37 UTC • •69.0

Article Content

1 / 4

Browse articles

ThreatCluster •August 8, 2026 at 20:35 UTC

•AI coding tools from Anthropic, Google, and OpenAI are vulnerable to GitHub issues.
•A critical CVSS score of 10.0 was assigned to Google's Gemini CLI vulnerability.
•Patches are available for Claude Code and Gemini CLI, but Codex requires workflow changes.

AI coding tools from Anthropic, Google, and OpenAI were found vulnerable to malicious GitHub issues. Researchers at Novee Security identified flaws in Claude Code, Gemini CLI, and Codex that could allow remote code execution, credential theft, and repository modifications. The vulnerabilities stem from hidden instructions in GitHub issues, with Gemini CLI receiving a critical CVSS score of 10.0. While patches have been issued for Claude Code and Gemini CLI, OpenAI's Codex requires workflow changes without a product update. Novee Security urged companies to review their workflows and limit token permissions due to similar configurations in over 100 public repositories. The vulnerabilities pose significant risks to developers and organizations using these tools.

Start a free Starter trial for enhanced analysis

Ask AI about this cluster

Answers cite the sources they use

Updated 97d ago How this analysis works

Timeline

2026-08-07

Vulnerabilities disclosed by Novee Security

Novee Security revealed flaws in Claude Code, Gemini CLI, and Codex that could lead to credential theft and code execution.

2026-08-07

Patches released for Claude Code and Gemini CLI

Both Anthropic and Google issued patches for their respective tools, addressing the identified vulnerabilities.

2026-08-07

OpenAI Codex vulnerability requires workflow changes

Unlike the other tools, Codex does not have a patch and requires users to modify their workflows to mitigate risks.

More articles in this cluster (3)

Threat Graph

11 entities, 12 inferred relationships (STIX 2.1).

The full threat graph for this incident is free in the 7-day Starter trial.

Create free account

No card · 30 seconds

See everything included

Technique (4)

Data Breach
Malware
T1059 - Command and Scripting Interpreter
T1567 - Exfiltration Over Web Service

Weakness (1)

CWE-200 - Exposure of Sensitive Information

Affected Platform (2)

GitHub
GitHub Actions Runner

Tool (4)

Claude Code
Codex
Gemini CLI
Hugging Face

Relationships

Data Breach targets GitHub
Data Breach targets GitHub Actions Runner
Malware targets GitHub
Malware targets GitHub Actions Runner
T1059 - Command and Scripting Interpreter targets GitHub
T1059 - Command and Scripting Interpreter targets GitHub Actions Runner
T1567 - Exfiltration Over Web Service targets GitHub
T1567 - Exfiltration Over Web Service targets GitHub Actions Runner
Claude Code related to CWE-200 - Exposure of Sensitive Information
Claude Code related to Codex
Claude Code related to Gemini CLI
Claude Code related to Hugging Face

Continue Reading

Critical RCE Vulnerability in Confection WordPress Plugin (CVE-2026-115252) A critical vulnerability, CVE-2026-115252, has been identified in the Confection WordPress Cable plugin, affecting versions 0.4.0 to 0.4.7. This flaw allows unauthenticated attackers to execute arbitrary PHP code on the server by writing unmodified settings into a PHP file without proper checks. The vulnerability has…

Critical Code Injection Vulnerability in Staggs Plugin for WordPress A critical code injection vulnerability, CVE-2026-78287, has been identified in the Staggs plugin for WordPress, affecting versions up to 3.3.0. This flaw allows unauthenticated attackers to execute arbitrary code remotely, posing severe risks to affected sites. The vulnerability stems from improper control over code…