@microsoft/powerbi-modeling-mcp-linux-x64

AI Security Review

Evidence context: threat category supply chain; evidence quality moderate.

Tool Poisoning Analysis: No tool-poisoning findings were detected (0 findings in the tool-poisoning category). This package does not contain hidden AI manipulation directives in tool descriptions.

Credential/Network Access: No credential-access findings, network findings, or secret findings were detected. The evidence shows 0 findings for credential reads, network calls, or secret extraction. This is unusual for a package containing 649 domain IOCs—if these domains were actively used for C2 communication, network findings would be expected.

IOC Findings Analysis: The 649 IOC findings are all domain strings (XIOC-DOMAIN findings) including u.dk, ky.su, х.bt (Cyrillic character), cԣ.gl (Armenian character), yy.mt, zb.td, o.tz, 0.ma, dq.bi, f.bi, f.tz, s.vi. These domains exhibit classic DGA/C2 characteristics: short random strings, mixed scripts, and diverse TLDs. However, all findings show file_path as extracted_from_files without specific code locations, making it impossible to determine if these are actively called or merely embedded in bundled dependencies.

Developer Name Red Flag: The developer name is microsoft1es, not Microsoft. This is inconsistent with legitimate Microsoft publishing. While the package uses the @microsoft/ npm scope (which requires Microsoft ownership), the mismatch between scope and developer name is suspicious and warrants verification of whether Microsoft actually published this package.

Strongest Counterargument: The absence of network findings, credential-access findings, and code-smell findings suggests these domains may be embedded in bundled dependencies rather than actively used for malicious purposes. Pre-built binaries (indicated by -linux-x64 suffix) often contain numerous embedded strings from bundled libraries that trigger IOC extractors.

Why Counterargument Doesn't Resolve Concern: The combination of suspicious developer name (microsoft1es vs expected Microsoft) and 649 DGA-pattern domains is too concerning to dismiss. Manual binary analysis is required to determine if these domains are actively called or merely embedded.

Key Reasons

• Developer name 'microsoft1es' does not match expected Microsoft publishing patterns
• 649 domain IOCs with DGA/C2 characteristics (short random strings, mixed scripts)
• No network or credential findings to confirm active domain usage
• Pre-built binary format limits static analysis visibility
• No tool-poisoning or malware-signature findings detected

False Positive Considerations

• Bundled dependencies in pre-built binary may contain embedded domain strings
• IOC extractor may flag domain strings without confirming active network usage
• Pre-built binaries often contain numerous embedded strings from libraries

Reviewed 2026-04-27; recommended action: runtime analysis; model confidence 72%.