Ports, Marine Terminals, and MTSA-Regulated Waterfront Facilities: The 2026 AI Physical Security Sector Playbook
How port authorities, terminal operators, and Facility Security Officers close the gap between the Facility Security Plan and the fence line with AI detection.
American waterfront security is regulated at the plan level and defeated at the perimeter level. Three numbers explain the gap.
Port security in the United States is one of the most heavily documented physical security regimes in existence. The Maritime Transportation Security Act of 2002 and its implementing regulations at 33 CFR Part 105 require roughly 3,200 waterfront facilities to maintain a Coast Guard-approved Facility Security Plan, designate a Facility Security Officer, conduct a Facility Security Assessment, and demonstrate the ability to escalate security measures across three MARSEC levels. On paper, no other commercial sector matches that compliance depth. In practice, the daily failure mode at a marine terminal is not a missing plan. It is a fence line no one is watching at 0300, a gate camera recording a trespasser rather than detecting one, and a container yard where a loaded trailer disappears between shift changes.
This sector playbook is for port authority security directors, terminal operators, Facility Security Officers, and the risk and insurance leaders who sit above them. It covers where AI physical security fits inside the MTSA compliance architecture, what the 2025 cargo theft data means for terminal threat models, how computer vision handles the specific optical conditions of waterfront environments, and how the FY2026 Port Security Grant Program treats AI video analytics as an allowable cost. The premise throughout: AI detection does not replace the Facility Security Plan. It closes the gap between what the plan promises and what the facility can actually see.
What 33 CFR Part 105 actually requires, and where the attention gap lives
Every MTSA-regulated facility operates under a three-document compliance chain: a Facility Security Assessment that identifies vulnerabilities, a Facility Security Plan that commits the facility to specific countermeasures, and Coast Guard verification that the plan is being executed. Under 33 CFR Part 105 Subpart B, the facility owner or operator must designate a Facility Security Officer in writing, ensure continuous communication capability between facility security personnel, interfacing vessels, and the cognizant Captain of the Port, and maintain the ability to notify facility personnel of changes in security conditions at any time.
The regulation is explicit about outcomes and largely silent about methods. Section 105.275 addresses security systems and equipment maintenance; the surrounding sections require facilities to control access, monitor restricted areas, and screen persons and vehicles at rates that escalate with MARSEC level. How a facility monitors a restricted area at 0300 on a Tuesday is left to the plan. Most plans answer with some combination of fencing, lighting, patrols, and CCTV, and this is where the compliance architecture meets the same attention problem documented across every other camera-dense sector.
A container terminal or bulk cargo facility routinely operates 100 to 400 cameras across gates, wharfs, container stacks, warehouse interiors, and miles of fence line. A security operations center watching rotating feeds observes a small fraction of that footage in real time, and vigilance research dating back to the Mackworth radar-operator studies shows human detection accuracy on static monitoring tasks degrades within 20 to 30 minutes. The result is a facility that is compliant on paper and forensic in practice: the cameras record the intrusion, the investigation reconstructs it, and the Facility Security Plan is technically satisfied the entire time. The Coast Guard's own inspection history, documented in GAO's review of MTSA facility compliance (GAO-08-12), found inspections routinely identify and correct deficiencies, but an inspection regime verifies the presence of countermeasures, not their moment-to-moment effectiveness.
Why TWIC's documented weaknesses argue for object-level detection, not more identity screening
The Transportation Worker Identification Credential was designed to keep threat actors out of secure port areas through biometric identity vetting. Two decades of oversight have questioned its effectiveness: GAO found the card-reader pilot results incomplete and unreliable (GAO-13-198), and a congressionally mandated RAND assessment (RR-3096) concluded DHS could not demonstrate the credential's security value over prior access-control approaches. The lesson for terminal operators is not that identity screening is worthless. It is that identity is the wrong layer for detecting behavior. A valid TWIC says nothing about whether its holder is loitering at a container stack at midnight, whether an unbadged person just crossed the waterside fence line, or whether a vehicle is staged at a gate it has no business entering. Object, posture, and zone-based computer vision watches behavior directly, without computing anyone's identity, which is also why it adds a detection layer without adding a biometric privacy liability.
The 2026 waterfront threat model: strategic cargo theft, perimeter intrusion, and the interior violence risk nobody budgets for
The dominant economic threat to marine terminals in 2026 is organized cargo theft, and the trend line is steep. Verisk CargoNet recorded 2,646 confirmed cargo theft incidents across the United States and Canada in 2025, an 18 percent increase over 2024, with estimated losses surging 60 percent to nearly $725 million. Average value per theft rose 36 percent to $273,990. CargoNet's analysis also documents a tactical shift: some organized groups are abandoning complex fraud schemes in favor of direct thefts of unattended, loaded trailers, exactly the asset class that accumulates in terminal yards and adjacent drayage lots. The geography matters for port operators specifically. The sharpest 2025 increase occurred in the New York City metropolitan area, where New Jersey incidents rose 110 percent, a corridor anchored by the largest port complex on the East Coast.
Perimeter intrusion is the second structural exposure. Marine terminals present some of the longest and least defensible fence lines in commercial security: miles of chain link crossing rail spurs, drainage easements, and waterside approaches that fencing cannot fully close. The intrusion window at a waterfront facility follows the same arithmetic documented in IntelliSee's perimeter intrusion analysis: the time between fence crossing and asset contact is typically measured in tens of seconds, which means detection value decays to zero unless the alert reaches a responder while the intruder is still in transit.
The third exposure is the one terminal budgets consistently underweight: interpersonal violence and workplace safety inside the gate. Ports are industrial workplaces with 24-hour shift patterns, longshore labor operating heavy equipment, truck drivers queuing under time pressure, and gatehouse staff handling access disputes face to face. Bureau of Labor Statistics occupational data places transportation and material moving occupations among the highest-fatality occupation groups year after year, and a fall, a struck-by incident, or an assault at a gatehouse carries the same OSHA General Duty Clause exposure at a terminal as it does in any other workplace. A detection platform deployed for cargo and perimeter protection covers this exposure on the same cameras at no marginal hardware cost, which changes the ROI arithmetic in a way single-purpose systems cannot match.
Mapping AI detection to the MARSEC escalation ladder
MARSEC levels are the operational spine of MTSA compliance, and they expose a structural weakness in manual security models: escalation multiplies workload precisely when staffing cannot multiply with it. At MARSEC 1, a facility executes its baseline plan. At MARSEC 2, 33 CFR 105 requires heightened monitoring, increased screening rates, and additional patrols of restricted areas. At MARSEC 3, the facility must be capable of near-continuous monitoring and maximum screening. A terminal that meets MARSEC 2 obligations by doubling patrol frequency is buying overtime; a terminal that meets them by tightening detection thresholds and expanding monitored zones in software is reconfiguring a system it already owns.
How AI detection absorbs the escalation burden 33 CFR 105 places on terminal staffing
Each MARSEC step multiplies monitoring obligations. Software escalates in minutes; patrol rosters escalate in overtime.
Baseline operations
Regulatory posture: execute the approved Facility Security Plan. Control access, monitor restricted areas, screen at baseline rates.
AI detection postureContinuous perimeter, gate, and yard monitoring across all cameras simultaneously. Trespass, loitering, vehicle, and weapon detection at standard confidence thresholds. Every alert timestamped for the compliance record.
Heightened risk
Regulatory posture: increased screening frequency, additional restricted-area patrols, heightened monitoring of waterside and shore approaches.
AI detection postureDetection zones expanded and sensitivity thresholds tightened in software within minutes. Loitering dwell-time triggers shortened. Alert routing broadened to additional responders. No new hardware, no new headcount.
Incident probable or occurred
Regulatory posture: maximum screening, near-continuous monitoring of the entire facility, coordination with the Captain of the Port and responders.
AI detection postureFull-facility zone coverage active at maximum sensitivity. Detections route simultaneously to the security console, mobile devices, and first-responder channels, giving the COTP coordination chain a live detection record.
The five detection modalities that matter on a working waterfront
Marine terminals benefit from a narrower, deeper detection set than mixed-use commercial facilities. Five modalities carry most of the operational value.
Detection Modalities Mapped to Terminal Zones and MTSA Obligations
| Detection Modality | Terminal Zones | MTSA / Operational Tie-In |
|---|---|---|
| Perimeter and Restricted-Zone Intrusion | Fence lines, waterside approaches, rail gates, secure yard boundaries | Directly supports 33 CFR 105 restricted-area monitoring obligations; converts recorded breaches into real-time alerts during the transit window |
| Loitering Detection | Gate queues, container stacks, chassis pools, warehouse doors | Surveillance-detection countermeasure; organized theft crews conduct reconnaissance before strikes, and dwell-time anomalies are the visible signature |
| Vehicle Detection | After-hours yard movement, unauthorized staging areas, closed gates | A vehicle inside a closed terminal outside operating hours is the highest-confidence theft precursor in the 2025 CargoNet tactical data |
| Drawn Firearm Detection | Gatehouses, admin buildings, labor halls, truck queues | Workplace violence coverage under the OSHA General Duty Clause; gate disputes and cargo hijacking attempts are armed-threat scenarios |
| Fall Detection | Wharfs, container stacks, warehouse floors, maintenance areas | 24-hour industrial workplace with lone-worker exposure; a detected fall on a night shift reaches a responder in seconds rather than at the next patrol pass |
The optical environment deserves honest treatment. Waterfronts combine fog, salt haze, glare off water, sodium-vapor and LED lighting mixtures, and extreme scene depth. These are exactly the degraded conditions examined in IntelliSee's technology briefing on occlusion, low light, and adversarial conditions: modern models trained on degraded imagery retain useful detection performance in these environments, but honest deployment planning treats camera placement, IR capability, and the DORI-standard pixel density thresholds covered in the camera requirements briefing as first-order engineering questions, not afterthoughts. A tuning period against site-specific conditions, marine layer mornings, crane shadows, gull traffic, is where false-positive rates are actually won.
How the deployment case differs across waterfront facility types
Container Terminals
The highest-value theft target and the largest camera estates. Detection priorities are after-hours vehicle movement in the stack, loitering at chassis pools, and perimeter intrusion along rail and waterside boundaries. Alert routing goes to terminal operations and security simultaneously, because a theft-in-progress is also a crane-operations safety issue. The adjacent drayage exposure is covered in depth in the cargo theft and yard security threat briefing.
Bulk and Breakbulk Terminals
Metal theft rose 77 percent in 2025 per CargoNet, driven by copper demand, and bulk terminals holding scrap, cathode, and concentrate inventories sit directly in that trend. Long, low-activity perimeters make continuous human monitoring economically impossible; zone-based intrusion and vehicle detection carry most of the value.
Cruise and Ferry Terminals
The only waterfront category where crowd dynamics dominate. 33 CFR 105 Subpart E imposes additional screening obligations on cruise ship terminals; drawn-weapon detection at embarkation halls, crowd formation alerts in queuing areas, and unattended-vehicle detection at drop-off lanes align with both the regulation and the mass-gathering exposure profile.
Petroleum, Chemical, and LNG Facilities
These CDC-handling facilities carry the most severe consequence profile and often dual regulation (MTSA plus CFATS-legacy or state equivalents). Perimeter intrusion detection and loitering alerts at waterside approaches feed directly into the transportation security incident prevention logic that justified MTSA in the first place. Deployment logic parallels the substation case in the critical infrastructure sector playbook.
Inland River Terminals
Barge fleeting areas and river terminals are the thinnest-staffed MTSA population, frequently a single guard or none overnight. AI monitoring functions as the force multiplier that makes 24-hour detection coverage possible at facilities that cannot justify 24-hour posts.
Shipyards and Repair Facilities
High-value work-in-progress, defense-adjacent contracts, and heavy hot-work safety exposure. Fall detection and restricted-zone monitoring around drydocks address the industrial safety file; perimeter and loitering detection address the security file. One platform, one camera estate, two budget lines served.
Paying for it: the FY2026 Port Security Grant Program and the cost-share calculus
The Port Security Grant Program is the sector's dedicated funding vehicle, and it treats surveillance enhancement as core allowable cost. FEMA opened $95 million in PSGP funding for FY2026, following $90 million in FY2025, to support risk-based protection of critical port infrastructure. Enhancement of maritime domain awareness and physical security capability, the categories AI video analytics falls under, have been consistent priority investment areas across recent cycles. Eligible applicants include port authorities, facility operators, and state and local agencies providing port security services.
Two practical notes for FSOs building an application. First, PSGP is a cost-share program (typically 25 percent non-federal match for private applicants), so the internal business case still has to stand on its own; the avoided-loss arithmetic from the cargo theft data and the staffing-efficiency case from the MARSEC mapping above are the load-bearing arguments. Second, competitive applications tie the requested technology to specific vulnerabilities identified in the Facility Security Assessment, which makes an AI detection proposal unusually easy to write: the FSA almost certainly already names the fence line, the yard, and the after-hours window as vulnerabilities. The broader federal funding landscape, including how PSGP interacts with other DHS preparedness grants, is covered in the grant funding intelligence briefing and the grant funding resource center.
What implementation looks like on a working terminal
The deployment model that works on a waterfront is the one that respects existing infrastructure. IntelliSee's architecture connects to the facility's existing camera network through the VMS already in place, with detection running on a dedicated on-premises appliance inside the terminal's own network. Video does not leave the facility for detection, no video is stored by the platform, and no facial recognition is performed, which keeps the deployment outside the biometric privacy liability perimeter that identity-based systems carry. Alerts route through existing dispatch consoles, to mobile devices, to radios, or through RapidSOS directly to first responders, sitting upstream of the facility's existing MTSA communication obligations rather than replacing them.
Initial detection coverage typically follows within days of appliance installation, with a one-to-two week tuning period against site-specific conditions. For the FSO, the compliance dividend is documentation: every detection, alert, and response timestamp becomes part of a defensible record that the monitoring commitments in the Facility Security Plan are being executed continuously, not just at patrol intervals. IntelliSee also holds DHS SAFETY Act Full Designation as a Qualified Anti-Terrorism Technology, which matters at MTSA facilities specifically because the statute's entire purpose is prevention of transportation security incidents; SAFETY Act protections attach to exactly that scenario class. How the platform works end to end is covered at how it works, and the wider industrial deployment context lives on the industries overview.
Frequently asked questions about AI physical security at ports and marine terminals
Does AI video detection satisfy any specific MTSA requirement?
MTSA and 33 CFR Part 105 are largely method-neutral: they require facilities to control access, monitor restricted areas, and escalate with MARSEC levels, but do not mandate specific technologies. AI detection supports those obligations by making continuous monitoring of restricted areas operationally real rather than nominal, and by generating a timestamped detection record an FSO can present during Coast Guard inspection. It is a means of executing the Facility Security Plan, not a substitute for it.
Can AI detection work with the cameras a terminal already owns?
In most cases, yes. The platform connects through the existing VMS to existing IP cameras; a rack-mounted appliance is installed on the facility network and no camera replacement or re-cabling is required. The engineering questions that matter are pixel density on the scenes that matter (the DORI standard is the reference framework) and IR capability for overnight zones.
How does AI detection perform in fog, glare, and marine weather?
Detection performance degrades gracefully rather than binarily in adverse conditions. Models trained on degraded imagery retain useful accuracy in fog, rain, and low light, and IR-capable cameras preserve overnight performance. Honest deployments treat the marine layer, water glare, and mixed lighting as tuning inputs during the calibration period rather than assuming laboratory conditions.
Does this involve facial recognition or TWIC identity data?
No. Detection is object, posture, and zone based: a person inside a restricted zone, a vehicle in a closed yard, a drawn firearm, a fallen worker. No facial recognition is performed, no identity is computed, and no biometric data is created, which keeps the system outside both the TWIC identity architecture and state biometric privacy statutes.
Is AI video analytics an allowable Port Security Grant Program expense?
Surveillance and physical security enhancement has been a consistent allowable and priority cost category in recent PSGP cycles, including the $95 million FY2026 opportunity. Competitive applications tie the technology to vulnerabilities already documented in the Facility Security Assessment. Applicants should confirm current-cycle priorities in the FY2026 Notice of Funding Opportunity.
What is the strongest economic argument for a terminal that already has guards and cameras?
Coverage arithmetic. A terminal with 200 cameras and a two-person overnight SOC is monitoring a low single-digit percentage of its footage in real time, while 2025 cargo theft losses rose 60 percent to nearly $725 million with an average loss approaching $274,000 per incident. One prevented trailer theft funds a material fraction of a deployment; the staffing efficiency during MARSEC escalations and the workplace safety coverage on the same cameras compound the return.
Continue the research
This playbook covers the terminal-level case for AI physical security inside the MTSA framework. For deeper reading on adjacent pieces:
- Cargo Theft and Distribution Center Yard Security, the threat intelligence briefing on strategic theft patterns upstream and downstream of the port gate.
- Transit, Rail, and Aviation sector playbook, the surface-transportation counterpart to this maritime analysis.
- Perimeter control solution page, the technical treatment of restricted-zone and intrusion detection.
- Request a structured risk assessment to map detection zones against your Facility Security Assessment.
More intelligence like this
New IntelliSee research drops monthly at most. Subscribe and get the next sector playbook, technology briefing, or threat intelligence report in your inbox the day it ships.
Request a Risk Assessment
Talk to an IntelliSee security specialist. No sales pitch — a structured conversation about your environment, your threat profile, and whether computer vision is the right fit.
Request a Risk Assessment