Threat actors are abusing the Model Context Protocol (MCP) to poison AI agents, exploit rug pulls, and trigger STDIO command injection across enterprise AI environments, with confirmed impacts on GitHub MCP, Clawdbot, Cursor, VS Code, Windsurf, Claude Code, Gemini-CLI, and other platforms. The report shows that these attacks often bypass SIEM and EDR detection by using legitimate tool calls and exposes multiple CVEs and MITRE ATT&CK techniques tied to malicious MCP servers, indirect prompt injection, and credential exfiltration. #ModelContextProtocol #Clawdbot #GitHubMCP #OXSecurity #Anthropic #Cursor #VSCode #Windsurf #ClaudeCode #GeminiCLI

New malicious npm packages have been found carrying a leaked Shai-Hulud clone that steals developer credentials, secrets, crypto wallet data, and account information. One of the packages also adds DDoS botnet functionality, and the activity is linked to the TeamPCP campaign and the suspected “phantom bot.” #ShaiHulud #TeamPCP #npm #OXsecurity #phantombot

The first analysis of software supply chain security based on the Open Software Supply Chain Attack Reference (OSC&R) threat framework has...