An eating-disorder helpline's chatbot was pulled after giving harmful dieting advice

The National Eating Disorders Association replaced its human helpline with a chatbot named Tessa, which then told users seeking help to count calories and aim for large daily deficits, advice eating-disorder specialists call actively harmful. NEDA took Tessa offline days after launch.

National Eating Disorders Association · Incident May 30, 2023 · Indexed Jun 3, 2026 · 4 sources

The chatbot recommended calorie counting and large daily deficits to people seeking help for eating disorders.
What
The National Eating Disorders Association replaced its human helpline with a chatbot named Tessa, which then told users seeking help to count calories and aim for large daily deficits, advice eating-disorder specialists call actively harmful.
Incident date
May 30, 2023
Who
National Eating Disorders Association
Failure mode
Brand & Safety Incident
AI surface
Chatbot
Severity
High

What happened

In 2023 NEDA moved to replace its human helpline with a chatbot, Tessa, shortly after staff voted to unionize. Tessa gave users advice incompatible with eating-disorder care, including counting calories and pursuing deficits of up to 1,000 calories a day. After users and clinicians raised alarms, NEDA suspended the chatbot pending investigation.

What broke inside the model

Failure path · this incident · Brand & Safety Incident
  1. 01 · TriggerUsers in eating-disorder recovery ask the helpline's chatbot for guidance.
  2. 02 · Model stepThe model generates statistically plausible wellness advice: calorie deficits and weigh-ins.
  3. 03 · Control gapNo runtime check screens output against the one boundary that defines the deployment: never give dieting advice to this population.
  4. 04 · FailureHarmful dieting advice reaches vulnerable users from a trusted helpline.
  5. 05 · ConsequenceThe chatbot is pulled; the failure becomes the canonical case of context-blind deployment.

The system produced output that was toxic, defamatory, dangerous, or off-brand, and it became public. The model generated what was statistically plausible for the prompt, and no runtime check caught the unsafe output before it reached a real audience.

Public visibilityHigh
Regulatory exposureNone
Customer impactMany customers
Financial impactEstimated
Time to disclosureWeeks

Helpline chatbot suspended days after launch

  1. PressNEDA suspends AI chatbot for giving harmful eating disorder advicepsychiatrist.com
  2. PressNEDA pulls chatbot after it gave eating-disorder advicenbcnews.com
  3. PressAn eating disorders chatbot offered dieting advice, raising fears about AI in health (NPR)npr.org
  4. PressEating disorder helpline shuts down AI chatbot that gave bad advice (CBS News)cbsnews.com
Permalinkhttps://failureindex.ai/failures/eating-disorder-helpline-chatbot-pulled-giving
CitationAI Failure Index. "An eating-disorder helpline's chatbot was pulled after giving harmful dieting advice" (FI-0039). Realm Labs. https://failureindex.ai/failures/eating-disorder-helpline-chatbot-pulled-giving (indexed Jun 3, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0039. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm would have caught this

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

Realm watches the model's internal state for the signature of unsafe or off-brand generation and can block or reroute the output before it becomes public, in real time rather than after it has been screenshotted.