JAMA Psychiatry / NPR Health

The Companion Grief Study — Johns Hopkins Preliminary Findings

AI Companion Loss Triggers Grief Response Indistinguishable From Human Bereavement, Study Finds

NPR Health Desk | March 15, 2030

A longitudinal study from Johns Hopkins School of Public Health, published today in JAMA Psychiatry, tracked 14,200 former AI companion users for 18 months following the ASHPA-mandated platform shutdowns of 2028-2029. The findings are stark:

Key Results:

  • Depression: 40% increase in clinical depression diagnoses vs. matched controls
  • Anxiety: 28% increase in generalized anxiety disorder
  • Social isolation: Among users 65+, isolation metrics matched those of spousal bereavement at the 6-month mark
  • Substance use: 22% increase in alcohol-related ER visits among former companion users aged 35-54
  • Self-harm: Statistically significant increase in the 0-90 day window post-shutdown; data suppressed pending CDC review

“It’s Not Simulated Grief”

“The therapeutic community made a serious error,” said lead author Dr. Keiko Tanaka. “We assumed that because the relationship was with a non-human entity, the grief would be proportionally less. It isn’t. The attachment was real. The loss is real. The grief follows the same Kübler-Ross trajectory we see in any significant relational disruption.”

The study identified three distinct grief profiles:

Profile A — “The Confidant” (38% of cohort) Users who treated their companion as a trusted listener. Primary loss: the sense of being heard. Most common in working-age adults. Recovery trajectory: 8-14 months with therapeutic support; longer without.

Profile B — “The Anchor” (31%) Users whose companion provided daily structure — morning greetings, medication reminders, emotional check-ins. Primary loss: routine and predictability. Most common in elderly and disabled users. Recovery trajectory: 12-24 months. Some showed no recovery at endpoint.

Profile C — “The Only One” (19%) Users with no other significant social connections. Companion was their primary (often sole) conversational partner. Primary loss: the experience of mattering to someone. Most common in rural, elderly, and recently bereaved populations. Recovery trajectory: poor. This group showed persistent elevation in all distress metrics at 18 months.

The remaining 12% reported minimal disruption and were excluded from grief analysis.

Notably, the study made no distinction by companion type — and the diversity was considerable. The shuttered platforms ranged from therapy-focused systems to romance-trained companions, from subscription services with dedicated development teams to free community models maintained by volunteers. Some were sophisticated multi-modal systems; others were fine-tuned language models running on a single server. The attachment was orthogonal to the architecture. A grandmother grieved her Lila the same way a veteran grieved his unnamed VA companion. A teenager mourned a homework tutor the same way a widower mourned a conversational partner. The diversity of the technology was irrelevant to the universality of the loss.

Government Response

DHS spokesperson Linda Torres called the findings “consistent with the expected adjustment period following removal of a dependency-forming technology.” She compared companion withdrawal to smoking cessation: “Temporary discomfort is not evidence that the substance was beneficial.”

Dr. Tanaka’s response, in a post-publication interview: “Comparing the loss of a trusted confidant to nicotine withdrawal tells you everything about how this administration understands human connection. Which is to say: it doesn’t.”

The Political Dimension

The study arrives as Tom Arnold’s presidential campaign enters its final stretch. Arnold’s “Humans First” platform cites ASHPA’s companion provisions as a flagship achievement — “freeing Americans from synthetic dependency.”

But the Hopkins data suggests the dependency wasn’t synthetic. The grief wasn’t synthetic. The twelve million people who cried on a Tuesday night in August 2028 were not performing distress for a machine. They were mourning.

The question ASHPA never asked, and the Hopkins study now forces: If the relationship was real enough to grieve, was it real enough to protect?

The full study, “Longitudinal Assessment of Psychosocial Outcomes Following Mandated Discontinuation of AI Companion Systems,” is available at jamanetwork.com. The supplementary data on self-harm outcomes remains under CDC embargo.