Research portal / Proposal
October 2026
Proposal
This research investigates how a risk-aware multi-agent AI system can monitor, coordinate, and automate routine private security operations while keeping human supervisors responsible for high-risk decisions.
Prepared by K.M. Fazle Rabbi, Department of Computer Science & Engineering, St. Cloud State University. Research advisor: Dr. Indira Dutta. The full source document is on the downloads page.
Problem
Private security operations produce schedules, locations, clock records, patrol checkpoints, incidents, qualifications, site rules, overtime, supervisor messages, and client requirements. Much of that information is already digital. Acting on it is still mostly manual. Supervisors, dispatchers, and managers notice the problem and coordinate the response.
A missed shift shows why this is not one task. The operation may need to:
- Detect the absence
- Determine whether the location is uncovered
- Contact the assigned officer
- Identify available replacement officers
- Verify qualifications
- Check overtime limits
- Consider travel distance
- Notify a supervisor
- Update the schedule and document the change
- Possibly notify the client
One event can touch several systems and several levels of risk. Rules can catch conditions they were written for. They do not, by themselves, decide which of these steps can be automatic and which require a person.
Gap
Research already covers multi-agent systems, autonomous agents, scheduling, human–automation collaboration, and security resource allocation. Security games have studied computational allocation of guards and patrols. Recent work has studied language-model agents that divide a task and communicate. Those lines are usually separate.
The open question is how coordination, operational data, workforce management, physical security work, and risk-based human oversight can sit in one framework for a private security organization. ASBOS does not try to remove people from that work. It asks which routine coordination can be delegated, and which decisions must stay with a person because a mistake could affect safety, employment, law, cost, or a client.
Objective and questions
Design and evaluate a multi-agent framework that can coordinate routine private security operations and change its level of autonomy with operational risk.
How can a risk-aware multi-agent AI system automate routine private security operations using real-time operational data while preserving human oversight for high-risk decisions?
- RQ1. Can specialized AI agents coordinate security operations more effectively than a single-agent or traditional rule-based automation system?
- RQ2. How should the level of AI autonomy change according to the operational risk of a decision?
- RQ3. How much human operational workload can be reduced without increasing incorrect or unsafe autonomous actions?
- RQ4. How effectively can a multi-agent system respond to interconnected events involving guard tracking, scheduling, patrol compliance, and incident management?
Specialist agents
Field Operations Agent
Monitors guard location, geofence status, clock-ins and clock-outs, patrol checkpoints, missed patrols, unexpected movement, and current shift activity.
Workforce Agent
Handles schedules, late arrivals, no-shows, replacement availability, overtime, qualifications, conflicts, and shift coverage.
Incident and Dispatch Agent
Handles incident reports, alarms, requests for assistance, supervisor notification, incomplete reports, follow-up, dispatch, and escalation.
Compliance and Business Operations Agent
Checks licenses, training, site requirements, working-hour limits, timesheet exceptions, overtime policy, and other organizational rules.
Coordinator
Supervisor and Coordinator Agent
Reads the other agents and decides among no action, an automated action, a recommendation, human approval, or immediate escalation.
Workflow
- Monitor
- Detect
- Reason
- Coordinate
- Act or Escalate
High-risk decisions remain under human control.
Agent duties
Duties specified for each role. The coordinator’s list is the set of outcomes it may choose after reading the others.
Field Operations Agent
- Guard GPS and geofence status
- Clock-ins and clock-outs
- Patrol checkpoints
- Missed patrols
- Unexpected location changes
- Shift activity and current guard status
Workforce Agent
- Employee schedules
- Late arrivals and no-shows
- Replacement officer availability
- Overtime
- Employee qualifications
- Scheduling conflicts and shift coverage
Incident and Dispatch Agent
- Incident reports and alarms
- Requests for assistance
- Supervisor notifications
- Missing incident information
- Follow-up activities
- Dispatch coordination and escalation
Compliance and Business Operations Agent
- Employee licenses
- Training requirements
- Site-specific requirements
- Working-hour restrictions
- Timesheet exceptions
- Overtime policies and organizational policies
Supervisor and Coordinator Agent
- No action
- An automated action
- A recommendation to a supervisor
- Human approval before execution
- Immediate human escalation
Autonomy
The levels adapt Parasuraman, Sheridan, and Wickens (2000) to this setting. Level 4 still leaves high-risk decisions with a person.
Level 0
Observation
The system monitors operations, identifies events, and records information. No operational decision is made automatically.
Level 1
Recommendation
The system detects a problem, analyzes the situation, and recommends an action. A human makes the final decision.
Level 2
Approval required
The system determines and prepares an action. A human supervisor must approve it before it is carried out.
Level 3
Routine autonomy
The system may carry out predefined low-risk operational actions. The action is logged and available for human review.
Level 4
Coordinated autonomy
Specialist agents may resolve an event together when it stays inside established policy and risk boundaries. High-risk decisions remain under human control.
Scenarios and method
The first prototype is planned around four cases: late arrival or no-show; guard leaves an assigned geofence; missed patrol checkpoint; incident with coverage impact. The protocols are on the experiments page. The comparison of a rule-based system, a single agent, and the multi-agent design is on the methodology page. Neither the prototype nor the comparison exists yet.
Scope
The initial study stays inside the workflow Monitor → Detect → Reason → Coordinate → Act or Escalate. It does not attempt to run an entire security company. Outside the initial scope:
- Payroll processing
- Recruiting
- Sales
- Invoicing
- Contract management
- Advanced video surveillance analysis
- Autonomous disciplinary decisions
- Complete company administration
Later questions, not part of the initial study:
- Automated client communication
- Predictive staffing
- Security resource optimization
- Automated report review
- Training and certification monitoring
- Equipment and asset management
- Operational risk forecasting
- Integration with video analytics
- Automated billing reconciliation
- Larger-scale coordination across a company
A longer-term question is whether a smaller management group could supervise a larger operation. That question is not part of the current design.
Expected contribution
If the study is carried through, it is expected to contribute a domain-specific multi-agent architecture, a risk-based account of autonomy, an event-driven way to coordinate the agents, test scenarios, a human-escalation framework, and a comparison with rule-based and single-agent alternatives. Those contributions are prospective.
Conclusion
Security operations are continuous monitoring, communication, and coordination. Digital systems for scheduling, reporting, and tracking still depend on people to notice a problem and organize the response. ASBOS proposes specialist agents that watch different parts of that work and collaborate when something breaks. The point is not to remove supervisors. It is to learn where autonomy is useful, where approval is still required, and how several agents can stay inside stated risk limits. The comparison is meant to produce evidence about benefits, risks, and limits. That evidence does not exist yet.
Human oversight
Guard tracking and workforce monitoring raise privacy and employment questions. Location should be collected only when the operation needs it, and a location anomaly should not be read as evidence of misconduct. Recommendations and automatic actions are to be logged so a supervisor can see what information was considered and why an action was proposed or taken.
Defined human responsibility matters when automated systems are used in operations (Tabassi, 2023). Recent work on agentic systems also argues for oversight through the workflow, not only after the fact (Dhanorkar, Passi, and Vorvoreanu, 2026).
The research prototype will not be allowed to independently make high-impact decisions:
- Employee termination
- Disciplinary action
- Hiring decisions
- Law-enforcement actions
- Accusations of misconduct
- Other significant employment decisions
The study is about operational coordination. It is not a proposal for automated punishment.
Estimated research budget
About $2,000. This is an estimate for designing a prototype and running tests, not a record of expenditure. Actual cost depends on the number of experiments, communication volume, model choice, cloud use, and how many scenarios are tested. Free tiers or university credits may be used. The estimate does not assume that every experiment can stay inside a free tier. The point of the testing budget is to include failure cases: a failed message, a delayed email, a retried webhook, an unanswered call, a missing GPS event, or several events at once.
| Resource | Purpose | Estimate |
|---|---|---|
| AI / LLM API usage | Development of specialist agents, coordinator reasoning, classification, and decision generation | $250 |
| Cloud hosting, database, storage, and background services | Application hosting, PostgreSQL, event processing, object storage, queues, and development infrastructure | $150 |
| Telnyx communication services | Development testing for SMS, numbers, voice, alerts, and webhooks | $100 |
| Resend email infrastructure | Transactional email, operational notifications, and delivery testing | $80 |
| Domains, DNS, and test communication configuration | Research domains, subdomains, email authentication, and isolated test environments | $40 |
| Maps, geocoding, and location services | Geofence testing, simulated locations, and distance calculation | $75 |
| Logging and monitoring | System logs, agent activity, error monitoring, and audit records | $50 |
| Synthetic data and automation tools | Simulated guards, shifts, incidents, patrols, schedules, and events | $50 |
| Development contingency | Unexpected API usage, service fees, or implementation requirements | $105 |
| Estimated research and development total | $900 | |
| Resource | Purpose | Estimate |
|---|---|---|
| Telnyx SMS and voice testing | Guard alerts, supervisor notifications, no-show workflows, escalation, voice events, and communication failures | $250 |
| Resend email testing | Operational email, supervisor and client notices, incident notices, and delivery events | $200 |
| AI / LLM experimental runs | Repeated single-agent and multi-agent scenarios under different conditions | $250 |
| Cloud compute, database, and infrastructure | Repeated experiments, event history, background jobs, and test environments | $150 |
| Location and geofence testing | Repeated GPS events, distance calculations, and geofence entry and exit | $75 |
| Load testing, queues, webhooks, and event simulation | Concurrent events, communication failures, retries, and multi-agent coordination | $75 |
| Test phone numbers, email identities, and test domains | Isolated communication endpoints for experiments | $50 |
| Data storage and experimental analysis | Experiment logs, agent decisions, measures, and evaluation records | $25 |
| Testing contingency | Additional usage and unexpected experimental requirements | $25 |
| Estimated testing and evaluation total | $1,100 | |
| Budget category | Estimate |
|---|---|
| Research and prototype development | $900 |
| Experimental testing and evaluation | $1,100 |
| Total estimated research budget | $2,000 |