CounselGuard

CounselGuard is a lightweight AI safety and assurance layer for legal practice. It addresses two critical moments in legal AI use: before information is submitted to an AI system, and after an AI-generated legal answer or draft is produced.

Before AI, CounselGuard performs an indicative pre-flight risk assessment across personal data, confidentiality, sensitive information, AI vendor/data handling, purpose, data minimisation and human oversight. It provides a risk level and practical controls such as redaction, minimisation, checking the AI environment and human review.

After AI, CounselGuard analyses an AI-generated legal output for signals that require independent verification, including excessive certainty, material legal claims, citations or authorities requiring verification, potential sensitive-information exposure and missing jurisdiction or context.

CounselGuard deliberately does not provide legal advice, determine legal compliance, or decide whether an AI-generated legal proposition is legally correct. It keeps the lawyer in the decision loop and captures a simple governance record of the review.

The project was built as an experimental working prototype to demonstrate how Responsible AI and AI Governance principles can be applied practically to everyday legal AI use.

The Key features of the Project are:

1. Before-AI Pre-Flight Risk Check — Assesses proposed AI inputs across personal data, confidentiality, sensitive information, AI vendor/data handling, purpose, data minimisation and human oversight, with an indicative risk score and recommended controls.

2. After-AI Output Assurance — Reviews AI-generated legal text for excessive certainty, claims requiring verification, citations/authorities requiring verification, potential sensitive-information exposure and missing jurisdiction/context.

3. Human-in-the-Loop Review — Clearly flags when human legal verification is required rather than presenting AI output as a final legal conclusion.

4. Actionable Risk Controls — Recommends practical actions such as redaction, data minimisation, checking the AI environment and independent verification.

5. Governance Record — Captures the review session with timestamp, matter, purpose and human-review status.

6. Responsible-AI Guardrails — Demo Mode provides indicative rule-based signals and explicitly avoids legal advice or legal-compliance determinations.