Knowledge Base Regulation Tracker Standards & Frameworks OSS Directory AI Incident Database Research Paper Library Case Studies Society Articles News Jobs About Charter Sign inBecome a Member
Home/AI Incident Database/Toxic agent flow: a public GitHub issue makes a coding agent leak private repository contents

Toxic agent flow: a public GitHub issue makes a coding agent leak private repository contents

An issue filed on a public repository carried instructions that an agent with the GitHub MCP server followed while triaging issues, pulling data from the user's private repositories and publishing it in a pull request on the public one.

Severity: MediumMCPToxic agent flowToken scopePrivate to publicLethal trifecta

Contribute

Spot an error, or know a record we should add? Members can propose additions and submit corrections.

Submit incidentReport a correction or request removal

Related in AI Incident Database

Zero-click prompt injection through a calendar invite hijacks an agentic browser

A crafted Google Calendar invite that looked routine carried instructions the Comet browser's agent followed when it processed the user's calendar, letting it read local files and exfiltrate data with no user…

Data exfiltration2026

Researchers' autonomous offensive agent gains full access to a consulting firm's internal AI platform in two hours

According to the security firm's disclosure, an autonomous agent pointed at the firm's internal generative AI platform with no credentials reached read and write access to its production database, including tens of…

Data exfiltration2026

ForcedLeak: indirect prompt injection exfiltrates CRM data from an enterprise agent

Instructions hidden in a web-to-lead form submission were later processed by an Agentforce agent, which followed them to query lead records and send the data to an attacker's domain that had been allowed by an…

Data exfiltration2025