Workplace Violence in America: The 2025 BLS Threat Intelligence Analysis of Incident Patterns, Industry Risk Concentration, and the Detection Gap

Every February, the Bureau of Labor Statistics publishes the prior year's Census of Fatal Occupational Injuries. In February 2026, that release confirmed what preliminary data had suggested: 470 American workers died from workplace violence in 2024, up from 458 in 2023 and representing 17.3% of all occupational fatalities that year. Of those 470 deaths, 379 involved firearms — 80.6% of all workplace homicides.

These are the numbers that appear in regulatory filings and press releases. They matter. But they represent only the most visible layer of a problem that runs substantially deeper and affects a far larger share of the workforce than fatal incident data alone suggests.

The BLS Survey of Occupational Injuries and Illnesses, released on the same annual cycle, documents 77,780 serious nonfatal workplace violence injuries in 2024 — injuries significant enough to require days away from work or restricted duty. That figure sits within a Bureau of Justice Statistics longitudinal estimate of 1.3 million nonfatal violent victimizations per year across the U.S. workforce, based on 2015–2019 averages compiled in partnership with the National Institute for Occupational Safety and Health. The ratio of serious nonfatal to fatal incidents across that window approaches 165:1. The 470 fatalities are the visible top of a substantially larger problem, and they are the incidents physical security architecture is least likely to prevent without dedicated detection infrastructure.

This report analyzes both BLS datasets in depth, identifies the industry and occupational risk concentrations the data reveals, examines what Bureau of Justice Statistics perpetrator typology research shows about how violence enters workplaces, and maps the detection gap — the structural mismatch between passive camera coverage and actual human monitoring capacity — that makes the statistical patterns directly relevant to AI-assisted detection platform decisions.

For the regulatory enforcement analysis of how OSHA's General Duty Clause applies to these statistics, see the companion Intelligence report. For the cost decomposition covering the full financial impact of serious incidents, see the seven-tier cost briefing. The present report focuses on the epidemiological layer: what the incident data reveals, where risk concentrates, and what that concentration implies for prevention architecture.

The Fatal Incident Record: What BLS CFOI 2024 Shows

The 2024 Census of Fatal Occupational Injuries, published February 19, 2026, recorded 470 workplace homicides — a 2.6% increase over the 458 recorded in 2023. Within that total, shootings accounted for 379 deaths, or 80.6% of all workplace homicides. Struck-by incidents, stabbings, and other physical assaults account for the remainder.

The 470 figure represents 17.3% of all 2,717 fatal occupational injuries recorded in 2024, making violence the second leading cause of occupational fatality behind transportation incidents. The workplace homicide rate across the full U.S. workforce reached 3.3 per 100,000 full-time equivalent workers — a rate that has fluctuated between 3.0 and 3.7 per 100,000 over the past decade without a sustained downward trend, distinguishing workplace homicide from the broader violent crime decline tracked in FBI data.

The gender distribution of workplace homicide is notably asymmetric in ways the aggregate figure obscures. Women represent 15.3% of all workplace violence fatalities despite constituting 47.4% of the workforce. This asymmetry reflects two intersecting factors: men dominate the highest-risk occupational categories (transportation, protective services, retail), and workplace homicides disproportionately reflect robbery-driven violence that falls more heavily on male workers in public-facing roles. But the female fatality profile differs structurally from the male one. Among women killed in workplace violence, a substantially higher proportion involves domestic violence that follows a victim into the workplace. Bureau of Justice Statistics analysis of perpetrator relationships estimates that 14 to 16% of female workplace homicides involve intimate partner or domestic perpetrators — violence originating outside the workplace that enters because the victim is reachable at a predictable location and time.

The industry distribution of fatal workplace violence concentrates in four sectors. Retail trade accounts for 24.6% of all workplace homicides — a figure reflecting the combination of cash-handling exposure, late-hour operations, isolated working conditions, and the physical layouts of small retail environments where egress and cover are limited. Transportation and warehousing ranks second, primarily through taxi driver, rideshare, and delivery driver homicides. Healthcare and social assistance ranks third on fatalities but first on nonfatal violence by a wide margin, a distinction examined in detail below. Government rounds out the top four, primarily through law enforcement and corrections worker incidents.

The BLS CFOI 2024 also recorded 733 total deaths from violent acts when self-inflicted violence in occupational settings is included alongside homicides. That broader figure encompasses the fatalities most relevant to behavioral threat assessment programs — programs designed to identify escalating risk before it reaches physical expression. The CISA March 2025 Pathway to Violence framework and the October 2024 U.S. Secret Service report on behavioral threat assessment in the workplace both document the consistent pattern that most serious Type III incidents (worker-on-worker violence) are preceded by observable behavioral signals that pre-incident detection programs can capture and act on before violence occurs.

The Nonfatal Violence Epidemic: BLS SOII 2024

Fatal incidents, despite their policy salience, are not where workforce violence primarily lives. The Bureau of Labor Statistics Survey of Occupational Injuries and Illnesses for 2024 recorded 77,780 serious nonfatal workplace violence injuries — cases involving intentional injury by another person that required days away from work, restricted duty, or job transfer. Of those, 54,230 resulted specifically in days away from work, representing injuries severe enough to remove workers from active employment temporarily.

The healthcare and social assistance sector accounts for the dominant share of that total, with a sector-wide violence injury rate of 17.1 per 10,000 full-time equivalent workers — a rate that exceeds the all-industry private-sector average by a substantial multiple and makes healthcare and social assistance an extreme statistical outlier rather than merely an elevated-risk sector. The rate compounds when broken down at the occupational level: psychiatric aides, who provide direct care in behavioral health settings, carry a violence injury rate of 543.6 per 10,000 full-time equivalent workers. No other major occupational category in the BLS dataset approaches this figure.

Extending the time horizon reveals the scale of the nonfatal problem more fully. A Bureau of Justice Statistics analysis conducted with NIOSH, covering 2015 through 2019, estimated an annual average of 1.3 million nonfatal violent victimizations against U.S. workers. Within that total, approximately 979,000 were classified as simple assaults and approximately 186,000 as aggravated assaults — attacks involving weapons, serious bodily injury, or threats with a weapon. The remaining incidents included robbery (approximately 101,000 annually) and rape or sexual assault (approximately 44,000 annually).

The gap between the BLS SOII serious injury figure (77,780 in 2024) and the BJS total victimization estimate (1.3 million annually) reflects methodological scope differences: BLS SOII captures employer-reported serious injury cases, while BJS uses victim-reported survey methodology that captures incidents falling below SOII reporting thresholds, including those not resulting in recordable injuries but still constituting criminal victimization. Together, the two sources frame a range — roughly 77,780 confirmed serious injuries at the low end and 1.3 million total victimizations at the high end — with the 165:1 ratio establishing why fatal incident data alone systematically underestimates the security challenge facing American workplaces.

Industry Risk Concentration: Where Workplace Violence Clusters

The BLS SOII data reveals a risk distribution far from uniform across the American economy. Workplace violence concentrates in specific industries and occupational categories with an intensity difficult to convey through fatality data alone, because the fatality record partially masks the exposure profile of the highest-risk sectors.

Healthcare and social assistance is the most extreme case. Its 17.1 serious violence injuries per 10,000 FTE far exceeds the all-industry private-sector baseline. Within healthcare, occupational granularity makes the picture more striking still. Psychiatric aides face a violence injury rate of 543.6 per 10,000 full-time equivalent workers — roughly one serious violence-related injury for every 18 full-time psychiatric aide positions per year. This figure is not a rounding artifact or a statistical anomaly. It reflects structural conditions that BLS data does not itself explain but that sector research has documented extensively: patients in acute psychiatric distress may be unable to regulate behavior in the way ordinary workplaces assume; the physical environments of behavioral health units — shared spaces, limited egress, the proximity requirements of direct care — create contact exposure that most workplaces do not replicate; and the therapeutic model requires workers to maintain close physical proximity to individuals who may be in the middle of psychotic episodes, acute suicidal crises, or severe personality dysregulation.

For a sector-specific analysis of how computer vision detection platforms are being deployed to address these structural conditions — and how they do so without facial recognition, video storage, or PHI exposure — see the Healthcare Workplace Violence AI Detection Playbook.

Beyond healthcare, the BLS data identifies several other elevated-risk sectors. Educational services records a serious violence injury rate of 8.4 per 10,000 FTE — more than double the all-industry private-sector baseline for those sectors, though substantially below healthcare. Protective service occupations — correctional officers, security guards — carry elevated rates that BLS reports separately from broader service-sector figures. Retail trade, while not the highest-rate industry on nonfatal violence, carries the highest proportion of fatal workplace violence incidents (24.6% of homicides) because the robbery-driven homicide exposure profile far exceeds its assault-rate ranking. The combination of cash handling, late-hour isolated operation, and physical accessibility creates a fatality risk that concentrates criminal-intent violence more than any other major industry.

LIVE CAM-12 · FACILITY ENTRANCE

Who Commits Workplace Violence: Perpetrator Analysis and Pathway Patterns

Understanding the BLS incident statistics requires layering in what Bureau of Justice Statistics and behavioral threat research reveal about perpetrator typology. OSHA's established typology categorizes workplace violence into four types based on perpetrator relationship to the workplace:

The typology has direct implications for detection architecture. Type I and Type III incidents are the categories most amenable to early visual detection. A drawn firearm approaching a facility entrance, a person loitering for an unusual duration near a staff-only access point, an unfamiliar individual moving through a restricted zone: these are the visual signatures that computer vision models trained on real-world threat scenarios identify within seconds. Type II violence, which dominates the nonfatal volume in healthcare, is structurally different because the perpetrator — the patient — is a legitimate occupant of the space. Here, detection value shifts toward environmental signals: zone violations (a patient outside authorized areas), crowd formation in waiting rooms, and behavioral escalation patterns that precede physical contact.

Behavioral threat research adds another layer. The U.S. Secret Service's October 2024 report, Behavioral Threat Assessment in the Workplace, and CISA's March 2025 Pathway to Violence framework document a consistent finding: Type III incidents — worker-on-worker attacks — are almost never sudden. They follow an identifiable pathway through observable behavioral precursors: escalating verbal threats, expressions of grievance against specific individuals, unusual interest in weapons, isolation, erratic behavior changes. Organizations with formal behavioral threat assessment units are statistically more likely to intervene before an incident escalates. The challenge is that most organizations lack these programs, and the behavioral signals are frequently attributed to workplace stress rather than an escalating threat pathway until after an incident has occurred.

The Detection Gap: Why Passive Surveillance Fails the BLS Risk Profile

The BLS data describes where workplace violence happens and in what volume. The detection gap describes why existing surveillance infrastructure is structurally unable to prevent most of it in real time, even when cameras are present and recording.

A typical large healthcare facility operates 200 to 600 surveillance cameras across its campus. A typical security operations center allocates two to four monitors to watch rotating feeds from those cameras in real time. Even with continuous, uninterrupted attention on every available monitor, the mathematical real-time coverage rate is 3 to 5% of the camera network at any given moment. The remaining 95 to 97% of footage is being recorded for forensic review after an incident occurs, not monitored for incident prevention.

This is the structural definition of passive surveillance: cameras are capturing, but no effective watch is occurring. The distinction matters because the BLS data describes incidents that developed in real time, in spaces with camera coverage. The cameras were present. The prevention was not.

The detection gap compounds with shift patterns, staffing shortages, and the increasing camera density of modern facilities. The number of cameras in a typical large hospital or commercial campus has grown substantially over the past decade, driven by hardware cost reductions, VMS capability expansion, and insurance requirements for documented coverage. Staffing growth has not kept pace. The result is a widening ratio of cameras to effective human monitors: more footage captured, proportionally less of it effectively watched in real time.

The statistical consequence is that the facilities with the most documented workplace violence risk — healthcare environments with 17.1 serious incidents per 10,000 FTE, behavioral health settings with psychiatric aide rates approaching 543.6 per 10,000 FTE — are also those most likely to have camera networks that outpace their effective human monitoring capacity. Camera density and prevention effectiveness are not the same variable, and the BLS data makes this structural mismatch consequential rather than academic.

There is a second dimension to the detection gap beyond attention limits: the specificity problem. Human monitors watching rotating feeds are generally trained to respond to incidents already in progress — a physical confrontation visible in the frame, a person down on the floor. The pre-incident behavioral signals that threat research identifies — drawn firearm approaching a perimeter, loitering near an access point, unusual movement through restricted areas — require specific, trained pattern recognition to detect at the visual level, particularly in low-light, crowded, or visually complex environments. This is precisely the pattern-recognition task that computer vision models are optimized to perform continuously, at scale, without the attention degradation that limits human monitoring.

What the BLS Data Reveals About AI-Assisted Detection Platforms

The workplace violence statistics in BLS data are not merely descriptive — they are analytically predictive about where prevention infrastructure will return the most measurable value. Three findings in the 2024 datasets map directly to the detection modalities that AI-powered computer vision platforms provide.

Firearm involvement in 80.6% of workplace homicides. If 379 of 470 workplace homicides in 2024 involved a firearm, the single highest-value pre-incident detection capability is drawn-weapon identification. A firearm becomes visible before it is used. Between the moment a weapon is drawn and the moment it is discharged, there is a window — measured in seconds — during which an alert could initiate a response that changes the outcome recorded in the CFOI data. AI gun detection running on existing cameras addresses this window directly. Detection to alert in under 30 seconds, routed to security dispatch and first responders, provides a material pre-incident response interval that passive surveillance cannot generate.

Healthcare's ~60% share of serious nonfatal violence. The concentration of nonfatal violence in healthcare is so extreme that a platform optimized for healthcare environments addresses the largest single slice of the total BLS nonfatal injury dataset. Perimeter and zone violation detection in behavioral health settings, loitering detection at emergency department entrances, and crowd escalation identification in waiting rooms address Type II violence — the category driving healthcare's nonfatal volume. These modalities do not require the perpetrator to have drawn a weapon; they identify the environmental and behavioral precursors that precede physical escalation.

Retail's 24.6% share of fatal violence. The robbery-driven homicide concentration in retail environments maps to drawn-weapon detection and unauthorized-access identification at point-of-sale areas, entrances, and after-hours perimeters. Late-hour retail environments with reduced staffing are exactly where the camera-to-monitor ratio is most unfavorable and where AI-assisted detection fills the monitoring gap most directly.

The platform architecture that addresses these three findings looks similar across environments: existing cameras connected to an on-premises detection appliance, pattern-specific models that do not rely on facial recognition or video storage, and alert routing to the responders who can act within the intervention window the data defines. For a detailed sector-specific treatment of how this architecture deploys in the highest-risk environment the BLS data identifies, see the Healthcare Workplace Violence AI Detection Playbook. For the active assailant briefing covering FBI and BLS fatality data in parallel, see the 2026 Active Assailant Threat Intelligence Briefing.

The BLS data also establishes the business case baseline for AI detection investments. As detailed in the seven-tier workplace violence cost decomposition, OSHA estimates workplace violence costs U.S. employers approximately $56 billion annually across direct medical, lost productivity, legal, and insurance exposure categories. Against the 77,780 serious nonfatal injuries the BLS SOII records — each representing not only a human cost but a direct employer liability — the per-incident economics of prevention versus post-incident response become clear.

Organizations evaluating AI detection for their specific facility type, camera count, and industry risk profile can begin with IntelliSee's structured risk assessment. The workplace violence tracker maintains current primary-source data for practitioners who monitor the statistics on an ongoing basis.

Frequently Asked Questions