The AI Message Review module applies the Fortress JMS artificial intelligence engine to inmate written communications — including electronic messages, kiosk emails, and grievance submissions — to surface content that may warrant intelligence review, even when it does not contain explicit keywords from the facility’s monitoring lists. This capability addresses one of the most persistent challenges in correctional communications monitoring: inmates who deliberately avoid flagged terminology while still communicating in ways that carry threat intelligence value. Rather than matching against a fixed list of terms, the AI model evaluates the semantic content, tone, intent signals, and structural patterns of each message, generating a risk score and a plain-English summary of why the communication was flagged for human review.
The AI Message Review queue displays flagged messages in order of AI-assigned risk score, with the highest-risk items at the top. Each queue entry shows the inmate name, message type, recipient (where applicable), date sent, AI risk score (displayed as a color-coded pill: green for low, amber for medium, red for high), and a brief AI-generated summary of the flag rationale. Intelligence officers open any message by clicking its row, which opens the full message text alongside the AI analysis panel. The AI panel presents the risk score, the specific passages that contributed most to the score (highlighted in the message text), the reasoning categories that were activated (e.g., coded language indicators, threat intent signals, external coordination patterns), and a confidence level for the overall assessment.
Officers review each message and make a disposition decision using the action buttons at the bottom of the review panel: Clear (no further action needed), Monitor (add to watch list without formal SAR), Create SAR (escalate to a Suspicious Activity Report), or Refer to Supervisor (flag for a second-level review without creating a formal report). Each disposition is recorded with the officer’s name, timestamp, and any notes entered in the disposition comment field. This documented review chain is essential for facilities that must demonstrate to oversight bodies that all flagged communications received a documented human review — not just automated detection — before a final determination was made.
The AI model is continuously refined based on disposition feedback from intelligence officers. When officers clear a message that the AI scored as high-risk, or escalate a message that the AI scored as low-risk, these corrections are fed back into the model’s training pipeline, progressively improving the accuracy and relevance of AI flags for the specific communication patterns present in that facility’s inmate population. Administrators can view model performance metrics — including flag rate, escalation rate, and officer override rate — in the AI Review configuration panel. This feedback loop, combined with the full audit trail of every AI flag and officer disposition, provides the documented accountability structure required when deploying AI-assisted monitoring tools in a regulated correctional environment.