1. Introduction
Active‑Session Safety Science (ASSS) proposes that safety states can be inferred from how a device is used—timing, rhythm, and interaction patterns—rather than what is said in its content. This literature review synthesizes five converging research pillars that collectively justify ASSS as a coherent scientific field:
behavioral signals in human–computer interaction (HCI)
coercive control and grooming
behavioral drift and baseline deviation
real‑time detection and layered defenses in safety‑critical systems
developmental privacy, autonomy, and identity formation
Together, these domains show that (1) internal states leak through behavior, (2) content can remain deceptively “normal” under coercion, (3) deviations from baseline are measurable, (4) safety‑critical systems rely on continuous, layered detection, and (5) surveillance harms the very developmental processes safety tools claim to protect.
2. Behavioral signals in HCI under stress and cognitive load
Work in behavioral biometrics and HCI consistently shows that stress, cognitive load, and divided attention alter interaction dynamics on digital devices. A systematic review of touchscreen interfaces highlights experiments using search‑and‑tap and cyclic tapping tasks under visual, manual, and cognitive load, noting that performance degrades as load increases—through slower taps, reduced accuracy, and lower throughput. One review notes that such tasks are used “under visual, manual, and cognitive load,” making them ideal for studying timing and rhythm changes in interaction.
More targeted studies quantify these effects in applied contexts. A 2025 distracted‑driving touchscreen study reports that increasing cognitive load while interacting with a touchscreen reduced pointing throughput by more than 58% and worsened overall performance. Another driving‑simulator study from the University of Washington found that touchscreen accuracy and speed declined under multitasking, and glance duration shortened under high cognitive load—directly implicating interaction rhythm and timing.
Stress and emotional arousal also leave measurable traces in touch behavior. Work on “stress from swipes” and mobile‑game swipe dynamics shows that emotional arousal changes movement dynamics and fine‑motor consistency, even when the task itself is simple. Affective‑touch research further demonstrates that stress and regulation are reflected in tactile interaction, though often via physiological and emotional measures rather than pure tap mechanics.
Taken together, this body of work supports a core ASSS claim: internal states such as stress, cognitive load, and arousal systematically alter tap timing, swipe velocity, motor precision, and interaction rhythm—even when the visible content of communication appears unchanged.
3. Coercive control, grooming, and the normal‑looking surface
Coercive control and grooming research explains why content alone is an unreliable indicator of safety. Peer‑reviewed work on coercive control in intimate partner violence describes it as a pattern that “reduces autonomy and liberty over time,” with victims adapting behavior to survive while outward communication can remain superficially intact. Reviews of coercive control link it to trauma, PTSD, and depression, emphasizing that it is often subtle, cumulative, and hard to detect from outside because it can resemble ordinary relationship conflict rather than overt violence.
Grooming research shows a similar mismatch between surface appearance and underlying harm. A study on sexual grooming stages found that participants could recognize grooming behaviors only imperfectly, supporting the idea that grooming often involves behavior that does not look obviously suspicious at first glance. Descriptions of grooming emphasize trust‑building, isolation, desensitization, and post‑abuse maintenance—processes that can preserve outwardly normal communication while progressively reshaping the victim’s choices and responses.
Across these literatures, a consistent theme emerges: “surface communication can remain plausible, ordinary, or hard to distinguish from normal interaction even while behavior and decisions are being constrained.” That is, there is a systematic mismatch between outward communication and inward autonomy. This directly motivates ASSS’s focus on behavioral signals rather than message content: when coercion and grooming are present, content lies, but behavior does not.
4. Behavioral drift, baseline deviation, and anomaly detection
A third pillar comes from research on behavioral drift, concept drift, and baseline deviation. In entangled human–AI interaction, recent work defines behavioral drift as a gradual, often unnoticed shift away from a baseline in judgment, decision‑making, verification behavior, and task delegation. The same work distinguishes cognitive drift (changes in beliefs, confidence thresholds, interpretive frames) from behavioral drift (changes in how and when people interact), and emphasizes that drift is incremental and hard to detect without longitudinal methods and process tracing.
In exploratory search, concept‑drift research shows that users’ information needs and relevance criteria change over time, causing their behavior to drift from earlier baselines. Interactive Bayesian approaches explicitly model and adapt to this drift, capturing how user behavior evolves across sessions.
Security‑oriented behavioral biometrics and anomaly‑detection work extend this logic to interaction telemetry. Multimodal behavioral profiling uses mouse dynamics and other signals to establish baselines and detect endpoint baseline deviation. Real‑time endpoint anomaly detection frameworks model normal telemetry and flag deviations using adaptive statistical methods, change‑point detection, and distance metrics.
Affective‑computing studies on stress detection use subject‑dependent biosignal features that express each individual’s deviation from their own baseline to detect stress and affect. This reinforces the idea that individualized baselines—and deviations from them—are more informative than absolute thresholds.
For ASSS, these findings justify continuous, individualized safety‑state scoring: humans exhibit stable interaction patterns, and gradual deviations from those baselines are detectable and meaningful. Behavioral drift and baseline deviation are not speculative—they are established, measurable phenomena.
5. Real‑time detection, layered defenses, and continuous monitoring
Safety‑critical systems in industry, healthcare, and infrastructure provide a mature template for how ASSS structures its architecture. Real‑time anomaly detection in industrial control systems (e.g., SCADA) uses sequence‑to‑sequence models and LSTM‑based autoencoders to detect data manipulation attacks as they occur. Embedded‑systems research describes self‑monitoring architectures where a real‑time operating system collects and timestamps events, feeding them to local AI/ML models for real‑time sequence analysis and proactive defense—explicitly emphasizing robustness, predictiveness, and local intelligence without cloud dependency.
Layered‑defense frameworks for safety‑critical systems stress fault containment and the elimination of single points of failure. Multiple layers are only effective if they provide complete, non‑single‑point‑of‑failure coverage of relevant faults, with design‑time avoidance, runtime detection, and fail‑safe mitigation working together. Layered test‑and‑evaluation frameworks for autonomous systems span planning, trajectory generation, and real‑time control layers, often wrapping autonomy features in safety filters to guarantee safe operation.
Continuous monitoring is increasingly recognized as essential for early detection and prevention. In patient safety, continuous vital‑sign monitoring with wearables detects deviations far more often than intermittent checks and is associated with improved outcomes and reduced workload. Biomedical real‑time monitoring in restricted environments is framed as a way to counteract human operator error due to inattention or fatigue.
ASSS and its first architecture, Active‑Session Defense (ASD), inherit these principles directly: real‑time, on‑device inference; continuous monitoring of the active session; and a defense‑in‑depth architecture with multiple independent behavioral engines and no single point of failure.
6. Developmental privacy, autonomy, and identity formation
The final pillar addresses why ASSS must be privacy‑preserving and non‑surveillance by design. Developmental psychology and family‑technology research show that digital surveillance—especially unilateral location tracking and control apps—can undermine autonomy, erode trust, and complicate identity formation.
A large study of digital location tracking found that roughly half of parents and adolescents report the use of tracking, and that tracking was associated with greater externalizing problems (aggressive and deviant behavior) and higher alcohol consumption, particularly for older adolescents and those experiencing lower positive parenting. The authors note that older adolescents increasingly strive for independence and autonomy, and that tracking may be perceived as controlling and intrusive, undermining autonomy and trust.
Reviews of parental monitoring styles distinguish autonomy‑supportive monitoring from controlling, restrictive monitoring. Autonomy‑supportive approaches—where adolescents have a voice in setting digital rules—are linked to more openness, less hiding, and better outcomes. Controlling surveillance, by contrast, is associated with secrecy, conflict, and poorer prosocial behavior. Children’s own perceptions of surveillance software are often sharply negative; many describe parental control apps as “stalking” and express anger that their privacy is invaded rather than their parents talking with them.
Broader analyses of “the monitored generation” argue that constant oversight reshapes concepts of trust, respect, and autonomy, and can stunt emotional development and problem‑solving skills. Continuous surveillance can produce learned helplessness, eroding motivation to take control of one’s digital identity and complicating the formation of a cohesive self, fragmenting “real” and online selves.
More general work on autonomy and identity in adolescence shows that emotional autonomy and identity commitment predict psychological wellbeing, while over‑controlling parental behaviors increase internalizing and externalizing problems. Digital‑age identity research emphasizes that online environments are powerful contexts for exploration, but that how adults structure oversight can either support or undermine healthy identity development.
For ASSS, these findings are decisive: any safety system that depends on content surveillance, unilateral tracking, or intrusive monitoring risks harming the developmental processes it claims to protect. A field that aims to protect children and vulnerable users must therefore prioritize privacy‑preserving, non‑content, behavior‑based approaches.
7. Synthesis: Why ASSS is scientifically necessary
Across these five pillars, a coherent picture emerges:
Internal states—stress, coercion, cognitive load, duress—reliably alter interaction behavior.
Coercive control and grooming often preserve normal‑looking communication while reshaping autonomy and decisions.
Behavioral drift and baseline deviation are measurable and can be detected in real time.
Safety‑critical domains rely on continuous monitoring and layered defenses to prevent catastrophic outcomes.
Surveillance‑heavy approaches undermine autonomy, trust, identity formation, and psychological wellbeing, especially in adolescents.
Active‑Session Safety Science integrates these findings into a single discipline: real‑time, non‑content, privacy‑preserving safety based on behavioral inference. The literature does not merely “inspire” ASSS—it requires it.