Youth suicide, inherited vulnerability, and the thin-skull shortcut that cannot replace proof
A young person’s suicide is a tragedy. It is not, by itself, proof against a technology company.
A screenshot is not a psychiatric history. A sequence of events is not necessarily a chain of causation. A corporation’s bank balance is not medical evidence.
My objection is to the shortcut: start with a death, identify a platform, and treat the rest of the young person’s life as background noise. That approach risks replacing an investigation with a target—and replacing prevention with blame.
Companies should answer for harm their wrongful conduct demonstrably caused. But neither grief nor anger supplies a missing causal connection. The serious question is not simply which app was used. It is what happened to this person, why, and what could have interrupted the progression toward death.
Investigate the life. Do not just prosecute the screen.
Start with the deaths—not the headlines
Dobin and colleagues’ September 2026 JAACAP review pooled 19 psychological-autopsy case-control samples: 958 suicide deaths and 1,102 community controls, aged 25 or younger. Approximately 70% were male. This is research on deaths, not merely unhappiness or hours online. The age range includes young adults, not only children.
| Factor reported in the abstract | Pooled odds ratio | 95% confidence interval |
|---|---|---|
| Suicidal ideation | 16.1 | 5.0–51.5 |
| Previous suicide attempt | 14.2 | 6.3–31.8 |
| Acute stressful life events | 13.7 | 9.6–19.5 |
| Mood disorders | 9.9 | 5.3–18.2 |
| Substance abuse, using the authors’ terminology | 8.9 | 6.1–12.9 |
These are the five strongest associations reported in the abstract. They are not absolute probabilities, independent causal shares, or percentages of responsibility for an individual death. The authors interpret the findings as suggesting a more proximal, impulse-driven risk profile. Their category of acute stressful events cannot be relabeled exclusively as relationship problems.
The practical question is what the records show before the death: prior attempts, psychiatric symptoms, substance use, immediate crises, and changes in functioning.
A reconstructed history must distinguish contemporaneous documentation from later inference. The fact that something was remembered after litigation began does not make it false, but its timing and corroboration deserve examination.
The child had a history before the account
Inherited vulnerability is not a footnote. O’Reilly and colleagues’ Swedish family study, involving approximately 2.76 million offspring, supported genetic and environmental contributions to the intergenerational transmission of suicidal behavior. Its genetically informed design also illustrates why an apparent environmental association may be confounded.
Genomic research supports a complex, polygenic contribution to suicidal behavior, with overlap—but not complete identity—with depression and other psychiatric traits. The work by Li and colleagues included suicide deaths and analyses accounting for genetic overlap with depression. It did not establish a single deterministic “suicide gene.”
The clinical investigation should therefore distinguish family history, inherited susceptibility, diagnosed illness, prior suicidal behavior, and documented deterioration. Those are different propositions, not interchangeable labels. The proposed review should ask what was already present, when it began, and how it was expressed.
Genetics belongs in the causal investigation. It is neither destiny nor automatic corporate immunity. Population heritability does not mean that a specified percentage of a particular death was caused by genes. Nor does it establish that the same death would have occurred at the same time without a disputed exposure. These studies provide no individual test of inevitability.
That distinction protects the defense from an avoidable error: using an important scientific finding to make an unsupported legal claim.
ADHD, anxiety, and impulsivity: developmental precision matters
In Sheftall and colleagues’ study, among suicide decedents with known mental-health problems, ADD/ADHD was recorded in 59.3% of children aged 5–11, versus 29.0% of those aged 12–14. Depression/dysthymia showed the reverse pattern: 33.3% versus 65.6%. These describe a subgroup of decedents; they do not identify the proportion of deaths caused by each diagnosis. They make overlooking ADHD in a younger child especially difficult to justify.
Impulsivity also requires precision. A longitudinal study of 10,286 young adolescents associated negative urgency—rash action under distress—and lack of perseverance with later suicidal thoughts and attempts. It did not establish that every measure of impulsivity was equivalent or that every death was unplanned.
Anxiety warrants investigation without exaggeration. A longitudinal meta-analysis found modest associations with subsequent suicidal thoughts and attempts, but no significant pooled association with suicide death. Its findings concerned the measured constructs and follow-up periods, not a conclusion that acute anxiety is irrelevant to every crisis.
Depression and substance misuse cannot be assigned exclusively to adults. Dobin’s youth review found strong associations for both and reported stronger substance- and alcohol-abuse associations than in the adult comparison meta-analysis.
Replacing “the app caused it” with “ADHD caused it” would repeat the same mistake with a different label. The organizing model must be multifactorial.
Relationships are not background noise
Sheftall’s study recorded family-or-friend relationship problems in 60.3% of childhood cases and 46.0% of early-adolescent cases with relevant circumstances available. These are descriptive findings, not relationship-attributable death fractions. They nevertheless support taking the interpersonal history seriously.
The proposed inquiry should examine conflict, rejection, loss, bullying, humiliation, school pressure, abuse, and changes in support.
Was there a crisis? Did it precede heavier online activity? Was the platform the source of harm, a channel through which a dispute unfolded, or a place where preexisting distress was disclosed?
Relationships and technology are not necessarily competing explanations. An online exchange may be part of a relationship crisis. The investigation must test that possibility, not sort the facts into convenient “offline” and “online” boxes.
The answer to scapegoating a company is not scapegoating a family. It is examining the evidence.
Ordinary viewing is weak evidence of individual causation
A 2024 JAMA Pediatrics meta-analysis covering 143 studies and more than one million adolescents found small average associations between social-media use and internalizing symptoms. Those are symptom associations, not estimates of suicide deaths caused by a platform.
Nesi and colleagues distinguished general use frequency from specific online experiences. The limited frequency studies did not show significant associations with the examined outcomes, whereas cybervictimization and problematic use were among exposures associated with self-injurious thoughts and behaviors. This is not a finding that every kind of video or interaction is harmless.
A 2025 JAMA cohort study of 4,285 young people similarly found that total screen time at baseline was not associated with the later outcomes examined, while high or increasing addictive-use trajectories were associated with suicidal thoughts or behaviors and poorer mental-health outcomes. The authors acknowledged that the observational design could not establish causation and that several psychosocial factors were not included.
There is adverse evidence a credible defense must confront. A separate longitudinal study of 11,876 young people found that within-person increases in social-media use preceded greater depressive symptoms. That finding matters; it does not by itself establish platform-specific causation of a suicide death.
The defensible argument is precise: viewing time alone is a weak substitute for a clinically grounded explanation of a death.
A nonsignificant average association does not prove zero risk. A small average association does not cap the possible effect of a particular harmful interaction. And an association involving addictive use cannot be generalized to every ordinary user.
The Dobin abstract does not quantify video viewing, social media, AI, or genetics. An exposure omitted from the abstract cannot be treated as a tested null effect. Its estimates also cannot be divided by a screen-time correlation to manufacture a numerical ranking of causes.
AI is a separate exposure—not a synonym for screen time
Consumer-chatbot testing involving simulated adolescent crises found safety shortcomings and differences between companion chatbots and general assistants. It tested responses under specified conditions; it did not measure a real-world suicide-death rate attributable to AI.
A companion conversation, an automated recommendation, and passive video viewing therefore require separate analysis. The reviewed evidence does not justify importing ordinary screen-time findings as an AI effect-size estimate.
For a case-specific opinion, I would ask what the authenticated exchange actually contained, which historical system generated it, what the user’s clinical course showed, and what mechanism is alleged.
Did the interaction intensify danger, discourage help-seeking, or disclose an existing crisis?
An alarming excerpt is a starting point for investigation, not a completed psychiatric explanation.
The thin-skull shortcut: an eggshell is not evidence
The eggshell-plaintiff rule addresses unusual susceptibility. In Benn v. Thomas, the Iowa Supreme Court required an appropriate instruction where evidence supported an accident-related contribution to death despite preexisting heart disease. The rule can affect proximate cause as well as damages; it is not simply a bookkeeping rule.
But susceptibility does not establish that the defendant inflicted the alleged injury.
The distinction is illustrated by Norman v. Greene, a September 2025 Pennsylvania Superior Court memorandum affirming dismissal where needed expert causation evidence was absent. The court explained that:
“the self-identified status as an eggshell plaintiff would not change the causation analysis.”
It is nonprecedential and was not a suicide case. Nevertheless, it illustrates the distinction between susceptibility and proof of the alleged aggravation.
That is the objection worth pressing: a vulnerable plaintiff is not a substitute for proof of a causal contribution. The eggshell rule cannot be used to answer the factual question by assuming the very aggravation in dispute.
Existing doctrine also recognizes a distinction between susceptibility and independently progressing damage. Koch v. United States discussed responsibility for additional harm and a possible reduction for deterioration that would inevitably have occurred without the accident. The government failed to prove the independent deterioration on that record. Burdens and apportionment rules must be checked under governing law.
A defendant should not be assigned unrelated losses merely because they appear in the same medical history. But alleging that a young person would have died anyway requires evidence, not a genetic label.
Biology does not divide into a courtroom pie chart
Consider a hypothetical in which a severe outcome occurs only when both a vulnerability and a trigger are present.
Remove the vulnerability and the outcome does not occur. Remove the trigger and it also does not occur. Both are necessary in that model; neither fact supplies an 80/20 allocation.
This is why evidence of inherited susceptibility does not logically exclude an external contribution—and why identifying a final external event does not erase the inherited history. The example is a causal model, not a conclusion about any individual case.
The eggshell rule is a rule of legal responsibility, not a scientific claim that biological vulnerability has no causal role. Scientific evidence can expose an unsupported application; it does not automatically invalidate the rule itself. Benn illustrates the legal distinction between an existing condition and an injury superimposed on it.
No fabricated genetic percentages. No presumed platform contribution. No predetermined exoneration.
Expert testimony must survive more than a dramatic timeline
Federal Rule of Evidence 702 requires sufficient facts, reliable methods, and reliable application. Its commentary addresses unjustified extrapolation and inadequate consideration of obvious alternatives. The 2023 amendment clarifies that foundation and application are not invariably questions only of evidentiary weight. Missing one article is not automatically disqualifying; the reliability of the actual opinion is what matters.
Heller v. Shaw Industries is particularly instructive. The Third Circuit upheld exclusion of central causation testimony heavily dependent on a flawed temporal relationship, while rejecting a requirement that a physician eliminate every possible alternative or always produce product-specific published research.
Thus, counsel should test unexplained transitions from distress to death, adults to children, or ordinary use to a specific harmful interaction. But a defense should not demand scientific certainty that the law does not require. Pennsylvania state courts retain Frye general acceptance under their own Rule 702; federal and state standards must not be conflated.
A chronology is evidence. An unexplained causal leap is not made reliable by an impressive résumé.
A focused due-process objection—not a constitutional slogan
The constitutional concern is strongest when directed at an arbitrary presumption or denial of a fair opportunity to contest causation. In Western & Atlantic Railroad v. Henderson, the Supreme Court invalidated the operation of a Georgia presumption concerning negligence and causation under the Fourteenth Amendment.
A proposed analogy would challenge an instruction that turns vulnerability plus death into presumed platform causation despite relevant contrary evidence. That is an argument about a specific procedural defect—not a holding that the eggshell doctrine is unconstitutional. Turnipseed, which upheld a different rebuttable presumption, shows why the actual operation and rational basis of the challenged rule matter.
Nor do punitive-damages decisions establish a general constitutional cap on compensatory liability for unexpectedly severe injury. BMW v. Gore concerned punitive excessiveness; Honda v. Oberg concerned judicial review of punitive awards. Neither declares the ordinary eggshell rule invalid.
The concrete challenge is to liability without the required proof—not to the existence of a vulnerable person.
Preserve the record and match the defense to the claim
Medical causation is not the entire lawsuit. Counsel must identify the defendant, product, challenged conduct, dates, governing law, and elements of the claim. Section 230 addresses treatment as publisher or speaker of another provider’s information, subject to statutory limits. It is not a universal exemption for a company’s own development of information.
Moody v. NetChoice recognizes First Amendment interests in editorial compilation. It does not decide every software-function or product-injury claim. These defenses require product- and claim-specific analysis, separate from a psychiatric opinion.
The proposed evidence review should preserve complete relevant exchanges, metadata, model or product versions, safety controls, and incident records. Federal Rule 37(e) addresses the loss of electronically stored information that should have been preserved. Selectively saving only favorable messages compromises the investigation.
Discovery of family and medical information should be lawful and proportionate, with suitable protective orders. Rule 26 also distinguishes expert-disclosure obligations and qualified protections for certain consulting work. Calling a widely distributed proposal “privileged” does not make it so.
Finally, later publications must not be confused with earlier notice. A paper published in September 2026 cannot, merely by appearing then, prove what a company actually knew before an earlier incident. Contemporaneous research and product-specific knowledge must be investigated separately.
Save lives while there is still a life to save
My prevention priorities are treatment, trusted human connection, protection during acute danger, restricted access to lethal means, and dependable follow-up. NIMH identifies listening, helping the person connect, reducing lethal-means access, and ongoing contact as components of prevention.
Treatment can make a measurable difference. A randomized trial of 173 high-risk adolescents found that dialectical behavior therapy reduced repeat attempts and self-harm more than the comparison treatment at the end of treatment. Between-group advantages weakened during follow-up, and the predominantly female sample limits generalization. This supports treatment—not a promise of near-total prevention.
Eyesight supervision means protective observation, not a vague instruction to “keep an eye on” someone. Where indicated, it should be continuous with the ability to intervene promptly. The Joint Commission requires one qualified staff member to one high-risk patient in clinical areas containing ligature or other safety hazards; monitoring in appropriately safer psychiatric areas is governed by organizational policy.
Trusted family members should be active partners. My clinical preference is to use safe, willing, capable relatives whenever home management is appropriate, with clear instructions, relief, and professional backup. The relevant question is whom the young person trusts and whether safety can actually be maintained—not whether the observer wears a uniform. NIMH incorporates family observations, trusted adults, home safety, and urgent evaluation when needed.
That preference is not a claim that family observation has been proved superior to trained staff. A relative who is unsafe, unavailable, or part of an abusive situation is not an appropriate safeguard. Family presence cannot replace required staff observation or necessary emergency care. Supervision must support treatment and reassessment, not substitute for them.
We do not need to wait for a breakthrough before applying existing preventive measures. But the evidence discussed here does not establish that current measures can eliminate nearly every suicide, or justify extrapolating suicide-prevention findings into a guarantee about homicide.
The goal is to prevent every death that can be prevented—not to promise an outcome the evidence cannot support.
Accountability requires evidence. Scapegoating requires only a target.
My concern is that an unsupported corporate-blame narrative may distract from the clinical vulnerabilities, acute crises, and practical protections that deserve attention. Investigating them does not excuse proven corporate wrongdoing. Investigating a product does not justify erasing the rest of the young person’s life.
A defensible expert does not begin by promising a result. The assignment is to identify what is documented, what is disputed, what is missing, and which causal explanation survives examination—including unfavorable facts.
A thin skull is not proof of a blow. A vulnerable mind is not proof of a platform’s causal responsibility.
Find the danger. Treat the illness. Protect the person. Preserve the evidence. Then decide responsibility on the record—not on the size of the defendant’s bank account.
Saving the next life matters more than finding the most convenient target for the last death.
