Duty to Warn

Abstract

The duty to warn, which MeSH classifies under confidentiality, is a clinician's legal and ethical obligation to breach a patient's confidence to protect an identifiable third party from that patient's threatened violence. It was established by the California Supreme Court in Tarasoff v. Regents of the University of California (1974, 1976), which converted a private therapeutic relationship into one carrying limited public responsibility. This article treats the duty as a problem in judgment under uncertainty: discharging it requires predicting rare, serious violence, and that low base rate makes even a careful risk judgment mostly false alarms. It traces the doctrine's origin, sets out the signal-detection and decision-threshold logic that governs when a warning is warranted, and surveys how clinicians discharge the duty across a fragmented legal landscape.

Keywords: duty to warn, Tarasoff, duty to protect, violence risk assessment, base rate

What the Duty to Warn Is

The duty to warn is the obligation, placed on a psychotherapist or other mental-health professional, to take reasonable steps to protect a third party when a patient presents a serious risk of violence toward that person. It is an exception to confidentiality, the default rule that what a patient tells a clinician stays between them; the duty holds that confidentiality ends where the public peril begins, so that a sufficiently serious and specific threat obliges the clinician to act to prevent it (Tarasoff v. Regents, 1976). MeSH files the duty under confidentiality precisely because it is the standing limit on that principle rather than a kind of it.

The phrase names two related obligations that the case law separated. A duty to warn is discharged by notifying the intended victim, or the police, of the danger; a duty to protect is the broader obligation to take whatever reasonable action the situation demands, of which warning is only one option — others include intensifying treatment, hospitalizing the patient, or notifying the authorities (Appelbaum, 1985). Most modern statutes frame the obligation as a duty to protect and treat warning as one permissible means of satisfying it.

Demo 1 — Why a Warning Is Usually a False Alarm

A clinic of 1,000 patients is sorted by a risk judgment into those flagged as dangerous and those cleared. Set the true base rate of serious violence and the judgment’s sensitivity and specificity, then read the positive predictive value — the chance a flagged patient really is dangerous — off the flagged column. Lowering the base rate is what collapses it.

Base rate of violence: 3%

Sensitivity (flags the truly violent): 0.60

Specificity (clears the non-violent): 0.80

Of 1,000 patients, 30 will be violent. The judgment flags 18 of them correctly but also flags 194 who will not be, so it issues 212 warnings. Its positive predictive value is 18 ÷ 212 = 8.5% — meaning 91.5% of the people warned about would not have been violent. At this low base rate the red false-alarm cell dwarfs the green one.

What makes the duty a topic in cognitive psychology, and not only in law and ethics, is that discharging it requires a prediction — the clinician must judge whether this patient will actually harm that person — and violence is a rare event that resists accurate prediction. The obligation is written as though the dangerous patient were identifiable in advance, but the science of risk assessment shows that separating the few who will act from the many who make threats and never do is a signal-detection problem with an unforgiving base rate. Figure 1 sets out that structure, which the rest of the article develops.

Historical Development

The duty originates in a single case. In 1969 Prosenjit Poddar, a student in psychotherapy at the University of California, Berkeley, told his psychologist that he intended to kill a young woman, Tatiana Tarasoff, who had rejected him. The therapist judged the threat serious and asked the campus police to detain Poddar, but no one warned Tarasoff or her family; two months later Poddar killed her. Her parents sued, and the California Supreme Court held that the therapists had a duty to have used reasonable care to protect the foreseeable victim (Tarasoff v. Regents, 1976).

The doctrine came in two rulings, and the distinction still organizes the field. The court's first decision, in 1974 (Tarasoff I), framed the obligation as a duty to warn the intended victim. After objections from the psychiatric and legal communities that a bare warning was often the wrong remedy, the court reheard the case and issued a second decision in 1976 (Tarasoff II) that recast the obligation more broadly as a duty to protect, dischargeable by whatever step the circumstances reasonably required (Buckner & Firestone, 2000). The court's own language — that the protective privilege of confidentiality ends where the public peril begins — became the doctrine's motto.

What followed was not a uniform national rule but a patchwork. Other state courts and legislatures adopted, modified, or declined the Tarasoff holding, so that by the time the duty had matured it varied sharply across jurisdictions: some states impose a mandatory duty, some make protective disclosure permissive, and a few reject it (Walcott et al., 2001; Kachigian & Felthous, 2004). A clinician's precise obligation now depends on where the clinician practices, a fragmentation later research has documented in detail (Johnson et al., 2014).

The Mechanism: Prediction Under a Low Base Rate

The duty presumes that a clinician can identify the patient who will be violent. Decades of research on violence risk assessment show that this identification is possible only in a probabilistic, error-prone way, because serious violence is a low-base-rate event and no assessment method — clinical or actuarial — is accurate enough to overcome that base rate cleanly (Monahan, 1993). The problem is the same one signal detection theory describes: any judgment that sorts patients into dangerous and not trades false negatives against false positives, and where the signal is rare, even a sensitive test flags far more people who will not offend than people who will.

The consequence is counterintuitive and central. Suppose a risk judgment correctly flags 60% of those who will become violent and correctly clears 80% of those who will not — figures at the optimistic edge of what the literature supports. If only 3% of the relevant population will actually be violent, then among every hundred patients the judgment flags, the great majority are false alarms, and its positive predictive value — the chance that a flagged patient really is dangerous — is under 10%. The worked example below computes this exactly. Improving the test helps only slowly; raising the base rate, by restricting attention to a genuinely high-risk subgroup, helps far more, which is why the doctrine's requirement of an identifiable victim and a specific threat is doing real epistemic work, not just legal line-drawing.

Demo 2 — When Does a Warning Become the Right Call?

Because prediction is uncertain, warning is a decision under uncertainty with the structure of a threshold. Warn when the estimated probability of violence exceeds p* = 1 ÷ (1 + R), where R is how many times worse an unwarned attack is than an unnecessary breach of confidentiality. Set the estimated risk and the cost ratio, and watch the threshold — and the recommended action — move.

Estimated probability of violence: 12%

Cost ratio R (unwarned attack vs. needless breach): 9×

The threshold is p* = 1 ÷ (1 + 9) = 10%. The estimated risk of 12% is above it, so the lower-expected-cost action is to warn. Judging an unwarned attack more costly (a larger R) lowers the threshold, warranting a warning at a smaller estimated risk; a lower base rate makes clearing any threshold harder.

Because prediction is uncertain, discharging the duty is a decision under uncertainty rather than a readout of a fact, and it has the structure of a threshold. The clinician must choose between two errors with very different costs: warning needlessly, which breaches confidentiality, can rupture the therapeutic alliance, and may stigmatize or frighten the patient and the supposed victim; and failing to warn when the patient does act, which is catastrophic. Formally, the expected-cost calculation says to warn when the estimated probability of violence exceeds a threshold set by the ratio of those two costs — when the harm of an unwarned attack is judged nine times the harm of an unnecessary breach, the threshold sits near a 10% estimated risk (Appelbaum, 1985). Modern practice replaces unaided clinical intuition with structured professional judgment, which disciplines but does not eliminate this reasoning (Skeem & Monahan, 2011).

Discharging the Duty in Practice

In practice a clinician confronts the duty as a graded set of options, not a binary of warn-or-not. Appelbaum's influential analysis broke the obligation into three tasks: assess whether the patient poses a serious risk of violence, identify who is endangered, and select and implement a protective action proportionate to the risk (Appelbaum, 1985). The protective action itself ranges along a ladder — continuing and intensifying treatment, warning the identifiable victim, notifying the police, or pursuing civil commitment — and the reasonable choice depends on how high and how imminent the assessed risk is.

Demo 3 — Proportionality: Matching the Action to the Risk

The duty is discharged not by a single act but by choosing a protective step proportionate to the assessed risk. Appelbaum’s ladder runs from intensifying treatment, through warning the victim and notifying police, to civil commitment — each rung trading more of the patient’s confidentiality or liberty for more protection of the third party. Slide the assessed risk and see the highest justified step.

Assessed risk and imminence: 30%

At an assessed risk of 30%, the highest proportionate step is Warn the identifiable victim — notify the specific person named or clearly implied as the target. Cost incurred: breach of confidentiality; possible alarm. A clinician may also take any lower step; the ladder sets the ceiling, not the floor.

What actually happens after a clinician acts is less studied than the doctrine that compels the action. When a warning goes to law enforcement, the response is uneven: a long-term follow-up of police officers found low familiarity with Tarasoff obligations but near-universal follow-up once a warning was received, most often by notifying other officers rather than by protective contact with the potential victim (Guina et al., 2022). The empirical case for the duty is likewise mixed: one econometric analysis found that the imposition of a duty was associated with a small increase, not decrease, in homicides, possibly because the erosion of confidentiality deters some dangerous individuals from seeking or speaking freely in treatment (Edwards, 2014).

The fragmented legal landscape compounds the clinical difficulty. Because the scope of the duty — whom it protects, what triggers it, and whether disclosure is mandatory or merely permitted — differs by state, and because it continues to be reshaped by new appellate decisions, a clinician cannot rely on a single national standard (Boulos et al., 2023; Radley & Felthous, 2025). Reviews of professional training have found that this variation is poorly conveyed to the clinicians expected to act on it (Johnson et al., 2014; Adi & Mathbout, 2018).

Table 1

The Clinician's Options for Discharging the Duty, Ordered by Assessed Risk

Note. The steps form a proportionality ladder: the higher and more imminent the assessed risk, the further down the clinician is justified in going, and each step trades more of the patient's confidentiality or liberty for more protection of the third party (Appelbaum, 1985).

Worked Example

The reason a warning is so often a false alarm can be made exact with the arithmetic of predictive value. Imagine a clinic that sees 1,000 patients in the relevant window, and suppose — generously — that a risk judgment has a sensitivity of 0.60 (it flags 60% of those who will truly become violent) and a specificity of 0.80 (it clears 80% of those who will not). Let the base rate of serious violence be 3%, so 30 of the 1,000 patients will actually be violent and 970 will not.

Apply the two rates. Of the 30 who will be violent, the judgment flags 0.60 × 30 = 18 (true positives) and misses 12 (false negatives). Of the 970 who will not, it wrongly flags 0.20 × 970 = 194 (false positives) and correctly clears 776 (true negatives). The clinician therefore issues a warning about 18 + 194 = 212 patients.

The positive predictive value is the fraction of those warnings that are correct: 18 ÷ 212 = 0.085, or about 8.5%. Put the other way, 91.5% of the patients flagged as dangerous would not have been violent. Raising the base rate is what changes this: if attention is restricted to a subgroup in which 20% will offend — an identifiable, specifically threatened target — the same test's predictive value climbs to about 43%. This is the quantitative core of the doctrine's demand for a specific, identifiable victim: it is the only lever that moves the base rate enough to make the prediction, and therefore the warning, defensible (Monahan, 1993; Skeem & Monahan, 2011).

Discussion

The duty to warn is best understood as the law imposing a decision rule on a problem the science calls hard. The Tarasoff court reasoned from a moral intuition — that a professional who learns of a lethal threat should not stand by — into a legal obligation to act on a prediction. Cognitive psychology supplies what the intuition omits: that the prediction is made under deep uncertainty, that the base rate of serious violence is low, and that any decision rule sensitive enough to catch the rare true threat will generate many false alarms. The doctrine's requirements of a serious threat, an identifiable victim, and reasonable proportionate action are, read through this lens, not merely legal hedges but the conditions under which the underlying prediction is accurate enough to justify overriding confidentiality.

The tension that results is genuine and unresolved. Confidentiality is not a courtesy but a working condition of psychotherapy: patients disclose dangerous thoughts because the room is private, and a duty that erodes that privacy may, at the margin, drive the most dangerous disclosures out of treatment altogether — the mechanism one empirical analysis invoked to explain why the duty did not reduce, and may have slightly increased, homicide (Edwards, 2014). Against that stands the identifiable victim who could have been warned. The doctrine is the legal system's attempt to price these two errors, and the threshold it sets is a value judgment dressed as a clinical one.

The boundaries matter as much as the principle. Tarasoff is a California decision, not a national rule, and the duty a clinician actually owes is a state-by-state mosaic that continues to shift (Kachigian & Felthous, 2004; Boulos et al., 2023). The prediction at its heart remains error-prone, and structured tools discipline the judgment without making it accurate in the way the law's language implies (Skeem & Monahan, 2011). The duty is real and enforceable, but it is discharged reasonably by acting well under uncertainty, not by predicting correctly.

Current Directions

The most active legal-scholarly work maps how far the duty has drifted from its origin. Recent analyses trace state-specific jurisprudence — what counts as a triggering threat, how identifiable a victim must be, and how appellate courts read the specificity requirement after new rulings such as Pennsylvania's Maas decision — showing a doctrine still being redefined a half-century on (Boulos et al., 2023; Radley & Felthous, 2025). The persistent finding across this work is that clinicians are asked to apply a rule whose content is genuinely unsettled and poorly taught (Adi & Mathbout, 2018; Johnson et al., 2014).

On the empirical side the priority is to learn what the duty actually accomplishes. The questions the field is beginning to ask — what police do with a warning, whether warnings prevent harm, and whether the erosion of confidentiality carries its own cost in deterred treatment — are the ones that would let the doctrine be evaluated rather than merely obeyed (Guina et al., 2022; Edwards, 2014). Running underneath both strands is the unchanged difficulty of violence prediction: as risk-assessment methods improve, the boundary of what a clinician can reasonably be expected to foresee will move, and the legal standard of reasonable care will move with it (Skeem & Monahan, 2011).

Common Misconceptions

The duty means a therapist must call the police whenever a patient expresses anger.

The duty is triggered only by a serious threat of violence against an identifiable victim, and it can be discharged by a range of proportionate steps — intensifying treatment, warning, or hospitalizing — of which calling the police is only one (Appelbaum, 1985).

Clinicians can reliably predict which patients will be violent.

Serious violence is a low-base-rate event, and even a good risk judgment yields mostly false positives; the predictive value of a dangerousness flag can fall below 10% (Monahan, 1993; Skeem & Monahan, 2011).

Tarasoff created a single nationwide duty to warn.

Tarasoff is a California ruling; other states adopted, modified, or rejected it, so the duty is mandatory in some jurisdictions, permissive in others, and absent in a few (Kachigian & Felthous, 2004; Boulos et al., 2023).

Glossary

Actuarial prediction.

Forecasting an outcome from a statistical formula applied to measured risk factors, as opposed to unaided clinical impression; more accurate on average but still bounded by the low base rate of violence.

Base rate.

The prevalence of an outcome in the relevant population; when the base rate of violence is low, even an accurate risk judgment produces mostly false positives, the fact that dominates the duty's prediction problem.

Confidentiality.

The default obligation that a clinician keep private what a patient discloses; the duty to warn is the standing exception to it, permitting or requiring disclosure to avert serious harm.

Dangerousness.

The judged likelihood that a person will commit future violence; the legal category the duty rests on and the psychological quantity that risk assessment tries, imperfectly, to estimate.

Decision threshold.

The estimated probability of violence at which warning becomes the lower-expected-cost action; set by the ratio of the cost of an unnecessary breach to the cost of an unwarned attack.

Duty to protect.

The broader obligation established by Tarasoff II to take whatever reasonable action the danger requires; warning the victim is one way to satisfy it, alongside treatment, notification, or commitment.

False positive.

A patient flagged as dangerous who would not in fact have been violent; the error the duty most readily produces, because a rare outcome makes false alarms outnumber true ones.

Foreseeability.

The legal test of whether a reasonable clinician should have anticipated the harm; the hinge on which liability turns, and the point at which the science of prediction meets the standard of care.

Identifiable victim.

A specific, readily ascertainable person the threat is directed at; the requirement that both narrows the duty legally and raises the base rate enough to make the prediction defensible.

Positive predictive value.

The probability that a patient flagged as dangerous truly is; computed as true positives over all positives, it falls below 10% when the base rate is low even for a fairly accurate judgment.

Privilege.

The patient's legal right to bar disclosure of therapeutic communications; the Tarasoff court held that this protective privilege ends where the public peril begins.

Reasonable care.

The standard the duty imposes: not a correct prediction but the diligence a competent clinician would exercise, judged by the appropriateness of the protective action rather than its outcome.

Signal detection theory.

The framework describing how a judge separates signal from noise by trading false positives against false negatives; the formal model of the clinician's warn-or-not decision.

Structured professional judgment.

A hybrid risk-assessment method in which the clinician rates a defined set of empirically supported risk factors but integrates them judgmentally; the current standard that disciplines, without eliminating, predictive uncertainty.

Tarasoff doctrine.

The body of law descending from Tarasoff v. Regents (1974, 1976) that imposes on clinicians a duty to warn or protect third parties endangered by a patient; adopted unevenly across US states.

Key Researchers

Paul S. Appelbaum (b. 1951). Columbia University psychiatrist whose analysis recast the obligation as a three-step clinical task — assess the risk, identify the endangered party, and select a proportionate protective action — the framework most clinicians are now taught. ORCID - Faculty Page

Alan R. Felthous (living). Forensic psychiatrist at Saint Louis University and long-time editor of Behavioral Sciences & the Law whose extensive analyses of state-by-state duty-to-warn jurisprudence map how far the doctrine varies across US courts. ORCID - Faculty Page

John Monahan (b. 1946). University of Virginia scholar and the foremost authority on the clinical prediction of violence, whose risk-containment guidelines translated the duty to protect into an operational clinical protocol. ORCID - Faculty Page

Debra A. Pinals (living). University of Michigan forensic psychiatrist working the psychiatry–law-enforcement interface, whose research examines what actually happens after a clinician issues a Tarasoff warning. Faculty Page

Dominic A. Sisti (living). University of Pennsylvania bioethicist who documented the interstate variation in Tarasoff-type duties and the gap it leaves in professional training. ORCID - Faculty Page

Jennifer L. Skeem (living). University of California, Berkeley psychologist and a leading figure in structured violence risk assessment, the empirical foundation on which any modern duty-to-protect judgment rests. ORCID - Faculty Page

Frequently Asked Questions

What is the duty to warn? It is a clinician's legal and ethical obligation to take reasonable steps to protect a third party when a patient poses a serious risk of violence toward that person, even if doing so requires breaching confidentiality (Tarasoff v. Regents, 1976).

What was the Tarasoff case? In 1969 a patient told his university psychologist he intended to kill Tatiana Tarasoff; no one warned her and he later did. Her family sued, and the California Supreme Court held that the therapists had a duty to use reasonable care to protect the foreseeable victim (Tarasoff v. Regents, 1976).

What is the difference between the duty to warn and the duty to protect? The duty to warn is discharged by notifying the victim or the police; the duty to protect is broader, satisfied by any reasonable action the danger requires, including treatment or hospitalization. Tarasoff II recast the obligation as the wider duty to protect (Appelbaum, 1985).

Does every US state have a duty to warn? No. Tarasoff is a California decision, and other states adopted, modified, or rejected it. The duty is mandatory in some states, permissive in others, and absent in a few, so a clinician's exact obligation depends on jurisdiction (Kachigian & Felthous, 2004; Johnson et al., 2014).

Can clinicians accurately predict violence? Only imprecisely. Serious violence is a low-base-rate event, so even a risk judgment with good sensitivity and specificity flags many people who will not offend; the predictive value of a dangerous-patient judgment can be under 10% (Monahan, 1993).

Why does the base rate matter so much? Because predictive value depends on it. When violence is rare, false positives vastly outnumber true positives among those flagged. Restricting the judgment to an identifiable, specifically threatened victim raises the base rate and is the main lever that makes a warning defensible (Skeem & Monahan, 2011).

Does the duty to warn actually prevent harm? The evidence is mixed. One econometric study found that imposing the duty was associated with a small increase in homicides, possibly because eroding confidentiality deters some dangerous people from seeking treatment, and studies of what police do with warnings show uneven follow-through (Edwards, 2014; Guina et al., 2022).

How does a clinician decide whether to warn? By assessing the seriousness and imminence of the risk, identifying who is endangered, and choosing a protective action proportionate to that risk, warning when the estimated probability of violence exceeds a threshold set by the relative costs of an unnecessary breach and an unwarned attack (Appelbaum, 1985).

References