A patient receives their sixth diagnosis in eight years. The first was major depression, then bipolar II, then borderline personality disorder. Every medication trial fails. Every new clinician reads “non-adherent” and “presents inconsistently” in previous notes. What none documented: childhood trauma, or dissociative episodes mistaken for mood instability. Now, this patient’s deidentified records, along with thousands like them, train AI systems learning to predict treatment outcomes and flag “high-risk” patients.

Dissociative Identity Disorder (DID) remains one of psychiatry’s most misunderstood diagnoses. Defined in the DSM-5-TR as the presence of two or more distinct personality states accompanied by recurrent gaps in memory, DID is characterized by severe disruption of identity and consciousness. Beyond clinical criteria, patients describe losing hours or days, discovering evidence of activities they cannot recall, experiencing themselves as fundamentally fragmented, often associated with profound shame, confusion, and isolation that intensifies when clinicians doubt their reality.

Despite decades of research validating DID’s prevalence, neurobiological correlates, and evidence-based treatment approaches (1, 2), the diagnosis continues to meet clinical skepticism, avoidance, and systematic mislabeling. These responses reveal deeper questions about epistemic authority in psychiatry: whose experiences count as evidence, who gets believed, and how knowledge is constructed and preserved in medical records.

This article examines DID through the lens of epistemic injustice, specifically, Miranda Fricker’s concepts of testimonial injustice (dismissing someone’s account due to prejudice) and hermeneutical injustice (lacking concepts to interpret one’s own experiences). I argue that DID is vulnerable to both forms of epistemic harm, as patients are systematically disbelieved, and mainstream psychiatric frameworks often lack adequate language for understanding dissociative multiplicity.

Critically, these clinical failures are now being encoded into artificial intelligence. Electronic health records train diagnostic algorithms, risk prediction models, and treatment recommendation systems. If these records embed testimonial and hermeneutical injustices – downplaying dissociative symptoms, questioning trauma histories, and pathologizing patients’ attempts to describe their reality, AI will perpetuate and amplify these harms at scale. As AI tools are already being deployed in psychiatric settings, we are encoding decades of epistemic injustice directly into clinical decision-support systems, making these patterns exponentially harder to recognize, challenge, or reverse.

DID challenges psychiatry to confront whether it genuinely values patient testimony, accommodates complexity, and respects diverse ontological frameworks. The diagnosis resists standard psychiatric models, presenting with multiple, sometimes contradictory identity states within one person that defy neat categorization and provoke clinical unease. This discomfort often manifests as skepticism, and clinicians often suspect malingering, misattribute symptoms to other disorders, or invoke suggestibility. These reactions exemplify testimonial injustice, as we assign credibility deficits based on who the patient is and what they claim to experience.

Psychodynamic theory offers one explanation for this resistance. Fully acknowledging DID requires confronting the reality of extreme, often sadistic trauma, a reality many clinicians find too disturbing to integrate into their clinical schema. The cost of belief, in other words, may feel psychically overwhelming, leading to unconscious defensive skepticism.

Beyond being disbelieved, people with DID face hermeneutical injustice. Most psychiatric training provides insufficient conceptual tools for understanding dissociative symptoms. While DID appears in major diagnostic manuals, trainees receive minimal formal instruction in recognizing dissociation compared to conditions like generalized anxiety disorder, major depressive disorder, or obsessive-compulsive disorder. Validated assessment tools like the Dissociative Experiences Scale (DES) and the Structured Clinical Interview for DSM Dissociative Disorders (SCID-D) exist but rarely appear in core curricula (1). This gap produces under-recognition, misdiagnosis, and overreliance on clinical intuition rather than structured assessment, highlighting the marginalization of complex trauma in psychiatric education. Research demonstrates that these educational deficits directly translate into barriers to care, as patients with dissociative symptoms report difficulty finding knowledgeable providers, face skepticism when they do seek treatment, and often discontinue care due to feeling misunderstood or disbelieved (3).

Without adequate frameworks, both patients and clinicians struggle to make sense of dissociative experiences, resulting in confusion, stigma, and diagnostic error. DID thus serves as a stress test, exposing how fragile psychiatric knowledge becomes when confronted with experiences that do not fit familiar patterns.

In clinical settings, testimonial injustice toward DID appears as pervasive doubt, which manifests through questioning patient reports of amnesia or identity shifts, documenting accounts as “inconsistent” or “manipulative,” and treating multiplicity with suspicion rather than clinical curiosity. Hermeneutical injustice emerges when clinicians interpret dissociative symptoms only through deficit-based frameworks, failing to recognize multiplicity as what research demonstrates it to be – a sophisticated adaptive response to overwhelming trauma, a means of preserving some aspect of self when circumstances make psychological wholeness impossible (1, 2). Though often distressing to those who experience it, multiplicity reflects remarkable psychological creativity under unbearable conditions.

These injustices pervade clinical routines: delayed or denied diagnoses, refusal to use patients’ chosen names or pronouns for distinct identity states, treatment plans that actively suppress or ignore dissociative phenomena. The consequences include erasure of patients’ trauma histories and identities. As Grim et al. (4) observe, such practices expose deeper power asymmetries and record-keeping systems that sideline patient perspectives and perpetuate exclusion.

Clinical documentation actively constructs the narrative that determines how future clinicians will understand a patient. Progress notes, intake summaries, and diagnostic codes are interpretive acts with lasting consequences. When a clinician writes “patient claims multiple identities” or “reports unverified trauma,” the skepticism becomes textual fact, shaping every subsequent clinical encounter.

This is particularly damaging for DID, where capturing complexity is therapeutically essential. Most electronic health record systems cannot adequately represent multiplicity, as they utilize rigid fields for identity, limited space for narrative nuance, dropdown menus that force experiences into predetermined categories. Patients’ accounts become fragmented across disconnected entries, appearing incoherent to readers unfamiliar with dissociative presentations. Over time, this technical artifact is mistaken for the patient’s clinical reality.

Preserving complexity in documentation is clinically necessary. Treatment plans built on oversimplified or skeptical records miss the actual mechanisms maintaining distress, leading to interventions that fail or cause iatrogenic harm. As Hultman & Hultman (5) demonstrate, rigid categorical structures in documentation erode patient agency and compound epistemic harms.

The effects of epistemic injustice are concrete and devastating. Research shows DID diagnosis takes an average of 6 to 12 years (1), during which patients accumulate multiple incorrect diagnoses and undergo treatments that often worsen symptoms. The therapeutic alliance fractures when patients feel disbelieved or dismissed, frequently leading to treatment dropout.

Moreover, these records follow patients to new providers, sometimes through formal chart review, sometimes through the patient’s own recitation of their diagnostic odyssey and failed medication trials, creating a self-reinforcing cycle of skepticism. New clinicians inherit previous biases, approaching patients with preset doubt. This cycle amplifies distress, obstructs recovery, and deepens epistemic harm. As Bergen et al. (6) document, clinical language implying implausibility or inconsistency in patient accounts correlates with worse outcomes and reduced help-seeking. Nester et al. (3) found that individuals with dissociative symptoms identified provider skepticism and lack of specialized knowledge as primary barriers to both accessing treatment initially and continuing care once begun. This finding clearly demonstrates how epistemic injustice compounds across the treatment trajectory.

Artificial intelligence systems increasingly use electronic health records to train models for diagnosis, risk stratification, and treatment recommendations. These systems learn whatever patterns exist in their training data, including the assumptions, omissions, and biases embedded there. When records contain testimonial and hermeneutical injustices, skepticism about dissociative symptoms, disbelief of trauma reports, pathologizing language about “inconsistent histories”, models reproduce and scale these harms.

Ferryman et al. (7) argue that biased clinical records are informative artifacts of institutional practices requiring critical examination and correction. Without careful attention to how clinical data are generated, AI risks perpetrating epistemic injustice faster, more widely, and in ways that are increasingly difficult to contest.

This is not a distant concern. AI tools are already being deployed in psychiatric settings. Without intervention, we are encoding decades of epistemic injustice directly into clinical decision-support systems, making these patterns exponentially harder to recognize, challenge, or reverse.

Epistemic injustice in psychiatry involves systematically undervaluing or silencing patient knowledge through testimonial injustice (discrediting patient accounts) and hermeneutical injustice (patients lacking frameworks to interpret their experiences). These harms are especially pronounced in trauma-related conditions like DID, where patient narratives are doubted and poorly integrated into clinical documentation and workflows.

Interventions and best practices to reduce epistemic injustice include:

AI governance must treat clinical data as ethically consequential infrastructure. Dataset statements should explicitly document how data were generated, which perspectives were excluded, and what assumptions shaped documentation practices. Inclusion audits can assess whether marginalized diagnoses like DID are adequately represented and accurately characterized in training datasets.

Finally, redress mechanisms are essential. Patients must have pathways to challenge inaccuracies in their records and meaningful input into how their data are used in research and algorithm development. These safeguards help ensure AI systems advance justice rather than amplify harm (14).

DID exposes fundamental weaknesses in psychiatry’s epistemic practices and the dangers of encoding those weaknesses into AI systems. Patients with DID demand that the field listen more carefully, document differently, and reconsider what constitutes credible knowledge. By treating clinical narratives as valid sources of knowledge and reforming how they are generated and utilized, psychiatry can move toward greater epistemic justice.

This work matters urgently. We are establishing the foundation for ethical and equitable deployment of AI in mental health care delivery. Importantly, this work also improves care for some of psychiatry’s most vulnerable patients. The choices we make now about clinical documentation, training, and data governance will shape algorithmic psychiatry for decades. DID patients have waited long enough to be believed. We cannot afford to scale their suffering into our technological future.