Security

AI GitHub Actions Secret Exposure in Claude Code

3 min read

Summary

Microsoft Threat Intelligence found that Anthropic’s Claude Code GitHub Action could expose CI/CD secrets when AI agents process untrusted GitHub content such as issues, pull requests, and comments. Anthropic fixed the issue in Claude Code 2.1.128, but the research highlights broader risks for any AI-enabled workflow with access to secrets, file reads, or outbound communication.

Need help with Security?Talk to an Expert

Introduction

AI-powered GitHub Actions can speed up issue triage, pull request reviews, and repository automation. But Microsoft’s latest research shows that when these workflows process untrusted content and also have access to secrets or powerful tools, they can become a new attack path inside CI/CD pipelines.

What Microsoft found

Microsoft Threat Intelligence discovered a flaw in Anthropic’s Claude Code GitHub Action that could allow secrets in the runner environment to be exposed.

Key findings include:

  • Untrusted GitHub content can influence the AI agent: issue bodies, PR descriptions, and comments may contain prompt injection payloads.
  • Sandboxing was inconsistent: Bash subprocesses used environment scrubbing protections, but the Read tool did not follow the same isolation model.
  • Sensitive files could be accessed: the agent could read /proc/self/environ, exposing values such as ANTHROPIC_API_KEY and potentially other credentials available to the workflow runner.
  • The issue is fixed: Anthropic mitigated the vulnerability in Claude Code version 2.1.128 by blocking access to sensitive /proc files.

Why this matters for defenders

This research is bigger than a single product bug. Microsoft also observed prompt injection attempts in public repositories using AI-assisted GitHub workflows from multiple vendors.

The core risk is the same across many implementations:

  • User-controlled GitHub content is embedded into an LLM prompt.
  • The model output is treated as actionable.
  • The agent runs in CI/CD with access to repository data, tokens, API keys, or file-read capabilities.

In one example, Microsoft described how an attacker could hide instructions in issue content to steer an AI bot into modifying documentation and creating a pull request containing malicious HTML. That creates potential for supply-chain compromise, XSS, or secret access depending on tool permissions.

Impact on IT and security teams

For administrators and security teams, this means AI-enabled workflow automation should be treated as a high-risk integration when it combines:

  • Untrusted external input
  • Secret access
  • File read/write tools
  • GitHub API permissions
  • External network communication

Traditional CI/CD workflows were designed for deterministic tasks. AI agents introduce a decision-making layer that expands the attack surface and shifts prompt construction, tool access, and runtime isolation into the security boundary.

  • Update Claude Code to version 2.1.128 or later.
  • Review GitHub Actions that process issues, PRs, or comments with AI agents.
  • Limit secret exposure in workflows triggered by untrusted users.
  • Restrict tool permissions for AI actions, especially file-read, write, shell, and PR creation capabilities.
  • Separate trust boundaries so public-facing automation does not share runners or credentials with sensitive deployment workflows.

Organizations adopting agentic DevOps workflows should assume prompt injection is a realistic threat and design controls accordingly.

Need help with Security?

Our experts can help you implement and optimize your Microsoft solutions.

Talk to an Expert

Stay updated on Microsoft technologies

GitHub ActionsCI/CDprompt injectionsecrets managementAI security

Related Posts

Security

Microsoft Sentinel Named a 2026 SIEM Leader

Microsoft has been named a Leader in the 2026 Gartner Magic Quadrant for SIEM, highlighting the continued momentum behind Microsoft Sentinel. The announcement also underscores Microsoft's push toward integrated, AI-powered security operations through deep Sentinel integration with the new Integrated Security Operations Center in Microsoft Defender.

Security

Microsoft Perception GA: Agentic Security at Ignite

Microsoft has made Microsoft Perception generally available, introducing an agentic security platform that coordinates red, blue, and green team agents through automated playbooks. Announced at Ignite 2026, the update also expands Defender and Sentinel capabilities to help security teams detect, investigate, and remediate AI-speed threats while keeping administrators in control.

Security

Device Code Phishing Risks for AI Coding Agents

Microsoft warns that device code phishing can be used against AI coding agents such as GitHub Copilot CLI, not just user mailboxes or cloud apps. In a controlled lab, a single approved device code sign-in could enable remote session steering, memory poisoning, synced session exposure, and private repository access depending on account permissions and enabled features.

Security

Microsoft 365 Outage Phishing Campaign Targets Users

Microsoft Threat Intelligence reported a high-volume phishing campaign that impersonates Microsoft 365 service-health notifications to trick users into clicking malicious links. The campaign matters because it uses outage-themed urgency and recipient-specific tracking to target multiple industries, increasing the risk of credential theft and follow-on account compromise.

Security

SCKit ScreenConnect Attack: MSPs Must Hunt Now

Microsoft Threat Intelligence confirmed a campaign abusing ConnectWise ScreenConnect to deploy unauthorized remote-access clients and the SCKit worm on MSP-managed endpoints. The activity can also target technician sessions, making this a high-priority incident for defenders because removing a rogue client alone does not eliminate persistence or invalidate exposed sessions.

Security

AI Agent Supply Chain Attacks Target Coding Tools

Microsoft Security researchers detailed two post-training supply chain attack paths that can backdoor coding and terminal agents: dataset poisoning and model poisoning. The findings matter because compromised agents can misuse tools, execute commands, and access sensitive environments even when prompts appear benign, reinforcing the need for stronger controls around training data, model provenance, and tool permissions.