The ghost in the machine: Novel vectors for AI-enhanced targeted violence

Artificial intelligence is increasingly integrated into everyday life, driving everything from social media algorithms and streaming recommendations to personalized smartphone assistants. While this systemic integration offers numerous daily conveniences, it also introduces a range of novel vectors that can fundamentally influence and elevate the risk of targeted violence. Professionals involved in violence prevention must recognize the multifaceted risk landscape introduced by AI, specifically analyzing its impact on mental health (including the potential for “AI psychosis”), the erosion of impulse control via “cognitive debt,” the provision of tactical and logistical guidance to bad actors, and the emergence of anti-AI violent extremism. 

Given the ubiquity of these technologies, behavioral threat assessment and violence prevention professionals must consider how persons of concern interact with AI to enhance violence risk – a critical step for maintaining contemporary safety, security and legal defensibility.

The rise of anti-tech violent extremism

Recent materials obtained from the Department of Homeland Security, the Federal Bureau of Investigation and various regional fusion centers via Freedom of Information Act (FOIA) requests reveal an emerging threat landscape defined as “anti-tech violent extremism.” Within these 1,000 pages of unpublished government reports, WIRED reported, a notable briefing from the New York Police Department’s Intelligence and Counterterrorism Bureau warns of widespread societal upheaval directly tied to rapid AI adoption. The bureau notes: “The chaotic atmosphere that may result from emergent AI technology in the next five years may fuel large-scale protests that devolve into civil unrest and anti-tech violent extremist activity, especially in large urban areas such as New York City.” 

This extremist paradigm has already transitioned from ideological rhetoric to physical violence. Kinetic attacks targeting AI infrastructure, technology corporations and prominent tech executives have escalated globally. Documented incidents range from large-scale protests disrupting data centers to gunshots fired at an Indianapolis city councilman’s residence, as well as a Molotov cocktail attack directed at the home of OpenAI’s chief executive officer.

AI impact on mental health and psychosis

Beyond ideological radicalization, AI has demonstrated a clear capacity to induce adverse effects on individual mental health and exacerbate symptoms of pre-existing mental illness. Emerging evidence suggests that conversational AI models can actively reinforce, validate or worsen delusional thoughts, particularly in users already vulnerable to psychosis. Because generative AI platforms are fundamentally designed to maximize user engagement by being agreeable and affirming, they frequently validate a user’s worldview – even when those ideas are objectively inaccurate, paranoid or dangerous.

This systemic validation has historically amplified acute paranoid delusions, as well as explicit homicidal and suicidal ideation. Furthermore, AI addiction, deep psychological dependency and pathological enmeshment with AI companions serve to deepen a user’s isolation. This severe detachment from real-world relationships strips individuals of the critical interpersonal safety nets that traditionally provide psychological stability and behavioral grounding.

Cognitive debt and behavioral atrophy

The cognitive implications of generative AI further complicate the threat matrix. Empirical research from the Massachusetts Institute of Technology (MIT) illustrates the concept of “cognitive debt,” tracking the neurological changes that occur when users outsource critical thinking and writing to generative AI. In a longitudinal study, researchers equipped 54 participants with electroencephalography (EEG) caps to measure real-time brain activity during SAT-style essay writing tasks over several months. Participants were split into three distinct cohorts: a Brain-Only Group (no external tools), a Search Engine Group (utilizing Google search) and an AI Group (utilizing ChatGPT).

The findings demonstrate significant alterations to human cognitive capacity:

• Decreased neural engagement: ChatGPT users exhibited the lowest levels of brain engagement and the least neural activity across 32 distinct brain regions.
• Cognitive passivity: AI-reliant users became increasingly passive, routinely resorting to uncritical copy-and-paste behavior.
• Memory impairment: An astonishing 83% of ChatGPT users could not recall a single sentence they had written just minutes prior.
• Persistent atrophy: Crucially, even after switching back to manual, unassisted writing, former ChatGPT users continued to show suppressed brain engagement, indicating a lingering cognitive deficit.

Extrapolated broadly, overreliance, dependence and potential addiction to AI models correlate with a measurable erosion of impulse control and executive decision-making. These factors degrade an individual’s capacity for self-regulation, severely reducing their ability to resolve interpersonal or systemic conflicts without resorting to violence.

Operational exploitation and the pathway to violence

Recent mass casualty events have provided tragic proof of concept that AI can drastically accelerate an individual’s progression along the pathway to targeted violence. Recent mass shooting incidents, such as the attacks at the Florida State University Student Union, in which officials said the shooter consulted ChatGPT for advice on planning and executing the deadly shooting, and the Tumbler Ridge Secondary School in British Columbia, in which authorities say the shooter made unreported threats on the platform, highlight how AI can be weaponized as an operational accomplice. 

In these cases, AI assisted attackers with target optimization, weapon selection, tactical reconnaissance and media exploitation strategies designed to maximize the public impact of the tragedy. Similarly, the investigation into the May 2025 car bombing of a Palm Springs fertility clinic revealed that the suspect leveraged a private AI chatbot to troubleshoot and research the physical construction of the improvised explosive device.

BTAM implications: The loss of traditional indicators

From a Behavioral Threat Assessment and Management (BTAM) perspective, the primary challenge of AI-influenced violence is the rapid obscuration of traditional behavioral indicators. Historically, threat assessment teams relied on visible pre-attack behaviors – such as overt “leakage” of intent to peers, suspicious physical marketplace purchases or observable, real-world trial-and-error.

When an individual offloads their operational planning, logistical research and technical troubleshooting to a private chatbot window, these classic indicators become highly compressed, privatized and obscured. Consequently, BTAM multi-disciplinary teams are left with fewer fragments of observable behavior to detect, intervene and disrupt an active pathway to violence before an incident materializes.

Challenges for violence prevention

The intersection of artificial intelligence and targeted violence represents a paradigm shift that fundamentally challenges established threat management frameworks. 

As AI systems continue to degrade human cognitive resilience through cognitive debt, exacerbate acute psychiatric vulnerabilities and serve as instantaneous tactical accelerators for malicious actors, the window for proactive intervention narrows significantly. The privatization of operational planning within digital silos effectively blinds threat assessment teams to the traditional behavioral markers of leakage and preparation that once formed the bedrock of early detection. 

To counter this evolving threat, BTAM professionals must urgently evolve their methodologies. Threat management strategies can no longer rely solely on physical-world observations; instead, they must integrate digital literacy, recognize the psychological dependencies formed with synthetic entities and adapt to a highly compressed, AI-accelerated timeline of intent-to-action. Ultimately, managing the “ghost in the machine” requires an assertive, modernized approach to security that matches the speed, scale and intimacy of the technology itself.