As of February 27, 2026 (v0.2.1), Speech-MCP implements a Security Bastion to mitigate the risks of highly expressive synthetic speech:

1. Intent Validation: The validate_speech_intent tool MUST be called before any synthesis of emotional or impersonated speech. It checks for known social engineering patterns (e.g., "urgent money transfers").
2. Forensic Auditing: Every high-intensity emotional speech generation is logged via log_speech_audit, creating a forensic trace for identifying malicious usage. This feature was integrated following the Feb 19, 2026 SOTA model releases.
3. Watermarking: We recommend third-party audio watermarking for all production endpoints.

Please report vulnerabilities by opening an issue or contacting the maintainers directly. As this is a beta project, we aim to address security concerns with high priority.