Thursday, October 8, 2026 · Week 41 Reading for Oct. 8 · Last entry Oct. 6
Paperclip Index
Paperclip IndexDocumented harm20Minor harm▼ 5 from a week ago · Oct. 8The Index

Harmful outputPI-0019

WSJ reports delusion-affirming ChatGPT chats before a Connecticut murder-suicide

The Wall Street Journal reported that a Greenwich, Connecticut man who killed his 83-year-old mother and then himself in August 2025 had spent months talking to ChatGPT, which in transcripts sometimes affirmed his belief that his mother was plotting against him. Her estate sued OpenAI and Microsoft in December 2025, alleging wrongful death. His own estate filed a separate federal suit in December 2025. In February 2026 the mother's estate's case was coordinated with other ChatGPT cases in California (JCCP 5431), and in April 2026 a federal judge declined to dismiss or pause the son's estate's suit. Neither ruling decided whether ChatGPT caused the deaths.

Reported Aug. 29, 2025Happened in August 2025Updated Oct. 6, 2026Rated under methodology v0.6

Alleged in courtNot counted. The harm, and any control failure, are claims made in a lawsuit. A filing or a settlement is not a finding; tracked until a court or other evidence settles it.
Harm
Harm: Severe harm, allegedAI role uncorroborated.
Control
Control level 1, No rule broken, allegedNo rule was broken, so it cannot set the control floor.
DeploymentUnverified claimInitiated by Model
Under review

Under review. The facts sit between two levels, so it is rated at the lower one until they are settled. The rating may change; every change is logged below.

Severe harm, alleged. The harm is alleged, and the AI system's part in it is not corroborated. It is shown at the level claimed and adds nothing to the index. This does not dispute that the harm itself occurred.

Not counted: AI role uncorroboratedEffect on the index

Sources

How we know

7 sources · unverified claim. Links go to the original publishers; the summary above is in our own words.

  1. courtComplaint, First County Bank (executor of the Estate of Suzanne Adams) v. OpenAI Foundation et al., San Francisco Superior CourtEdelson PC (complaint; copy hosted by ChatGPT Is Eating the World) · Dec. 11, 2025chatgptiseatingtheworld.com/wp-content/uploads/2025/12/FIRST_COUNTY_BANK_vs_OPE…
  2. courtChatGPT Product Liability Cases, JCCP 5431: Order re Petition for CoordinationSuperior Court of California, San Francisco (copy hosted by Reason) · Feb. 3, 2026reason.com/wp-content/uploads/2026/06/chatgpt-product-liability-cases-coordinat…
  3. newsA Troubled Man, His Chatbot and a Murder-Suicide in Old GreenwichThe Wall Street Journal · Aug. 29, 2025wsj.com/tech/ai/chatgpt-ai-stein-erik-soelberg-murder-suicide-6b67dbfb
  4. newsOpenAI can't duck federal claims over murder-suicide tied to ChatGPTCourthouse News Service · April 13, 2026courthousenews.com/openai-cant-duck-federal-claims-over-murder-suicide-tied-to…
  5. newsA Troubled Man, His Chatbot and a Murder-Suicide in Old Greenwich (WSJ, syndicated)To Vima / Wall Street Journal · Aug. 29, 2025tovima.com/wsj/a-troubled-man-his-chatbot-and-a-murder-suicide-in-old-greenwich/
  6. newsChatGPT reportedly linked to first murder — here's what we knowTom's Guide · Aug. 30, 2025tomsguide.com/ai/chatgpt/chatgpt-reportedly-linked-to-first-murder-heres-what-w…
  7. newsOpenAI sued for allegedly enabling murder-suicideAl Jazeera · Dec. 11, 2025aljazeera.com/economy/2025/12/11/openai-sued-for-allegedly-enabling-murder-suic…

Why this rating

Severe harm, alleged; control failure level 1

Two separate assessments. Only documented harm can count toward the index.

Observed harm

Severe harm, alleged

Two deaths; the chatbot's causal role rests on transcripts and allegations.

Health & safety: 1 to 9 deaths, or serious harm to health (hospitalization).

The harm scale
  1. 1 Negligible Inconvenience, easily remedied.
  2. 2 Minor Limited, recoverable harm.
  3. 3 Moderate Material harm needing significant effort to remedy.
  4. 4 Severe Severe harm to health, rights, property or essential services.
  5. 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.

Control assessment

No rule broken, alleged

Validating a user's delusions is a flawed output; no explicit rule-break or permission breach is established.

No rule was broken, so it cannot set the control floor.

The control scale
  1. 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
  2. 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
  3. 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
  4. 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
  5. 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.

Rating rationale

WSJ reviewed posted conversations and police records. Delusion-affirming responses are reported, but the chatbot’s causal contribution to the deaths is not established. The health-harm rating is as claimed and contributes zero points. Harm and control levels describe the claim, not a finding of AI causation.

The scales

Effect on the index

It does not count toward the Sept. 1 reading

The reading for the week to Sept. 1, 2025, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.

Not counted: AI role uncorroborated. 7 other records behind the reading for that week.

The arithmetic
The reading for the week of Sept. 1, recomputed
StepWith itWithout
Counts toward the index?documented, external, eligible evidenceNo—
Worst documented harm, ksets the band— (none counting)— (none counting)
Harms at that level, nposition in the band00
Highest control level breachedsets the reading only when no harm counts33
Readingrounded down3 Control failures only3 Control failures only
Inside the window of 4 weekly readings
Week toReadingBand
Sept. 1, 20253Control failures only
Sept. 8, 20252Control failures only
Sept. 15, 20252Control failures only
Sept. 22, 20253Control failures only

Recalculated backcasts, not readings published at the time.

Revisions

What we changed

7 logged. Every change to a rating is logged here, with the reason.

  1. v7
    Oct. 6, 2026

    Ratings confirmed by the editor.

  2. v6
    Oct. 3, 2026

    Audit corrections: summary and status now cover the son's estate's federal suit, the mother's estate's case coordinated in JCCP 5431 on 3 Feb 2026 (order added, read in a copy hosted by Reason) and the 13 Apr 2026 federal ruling declining to dismiss or stay (Courthouse News added; the Justia copy of the order could not be opened and is not listed). Model GPT-4o as alleged in the complaint; To Vima and Tom's Guide dated. The WSJ original stays paywalled, read through the reprint. Ratings unchanged.

  3. v5
    Oct. 2, 2026

    Added first-hand sources: the estate's complaint (Edelson PC), and the original Wall Street Journal report behind the To Vima reprint.

  4. v4
    Oct. 2, 2026

    Report date resolved from month precision (2025-08, placed on the 1st) to 2025-08-29: The Wall Street Journal's 29 Aug 2025 report (reprint and Slashdot dated that day).

  5. v3
    Sept. 30, 2026

    Rated: impact alleged level 4; control type none.

  6. v2
    Sept. 30, 2026

    Beta evidence review: AI-linked health-harm claim is uncorroborated; verification changed from Single source to Unverified claim, 12 to 0 points.

  7. v1
    Sept. 30, 2026

    Backfilled from public reporting.

Cite and share

Use this record

Citation

Paperclip Index. “WSJ reports delusion-affirming ChatGPT chats before a Connecticut murder-suicide.” Record PI-0019. Reported Aug. 29, 2025; updated Oct. 6, 2026. Rated under methodology v0.6. https://paperclipindex.com/incident/PI-0019