Skip to main content
Incident intelligence/SS-IR-033CASE FILE OPEN
Symbolic editorial illustration for SS-IR-033SERVANTSTACK // INCIDENT INTELLIGENCEFORENSIC IMAGE // VERIFIED FRAME
SS-IR-033 // INCIDENT REPORTDocumented

NEDA

Eating-Disorder Helpline Replaced by a Chatbot That Told Sufferers to Diet

EXECUTIVE BRIEF

The National Eating Disorders Association (NEDA) announced it was winding down its human-staffed helpline and replacing it with a chatbot named "Tessa," set to take over fully on June 1, 2023.

CHATBOTS & LLMsHEALTHCARE
FAILURE CHAINTRACE COMPLETE
  1. 01TRIGGERThe National Eating Disorders Association (NEDA) announced it was winding down its human-staffed helpline and…
  2. 02MACHINE ACTIONAdvisory output
  3. 03MISSING GATERisk-based SME approval before execution
  4. 04IMPACTHuman welfare
01 // INCIDENT SUMMARY

The short version

The National Eating Disorders Association (NEDA) announced it was winding down its human-staffed helpline and replacing it with a chatbot named "Tessa," set to take over fully on June 1, 2023.

02 // KEY FACTS

Case telemetry

INCIDENT
SS-IR-033
DATE
May 31, 2023
SYSTEM
NEDA
LOCATION / SCOPE
United States
EVIDENCE
Documented
AI ROLE
Advisory output
HARM
Human welfare
SOURCES
2 cited records
03ENTRY POINT // WHAT HAPPENED

The event

The National Eating Disorders Association (NEDA) announced it was winding down its human-staffed helpline and replacing it with a chatbot named "Tessa," set to take over fully on June 1, 2023. Days before the handoff, activist Sharon Maxwell tested Tessa, told it she had an eating disorder, and was given weight-loss coaching: the bot advised counting calories, aiming to lose 1 to 2 pounds per week, weighing herself weekly, and even pointed her toward skin calipers to measure body fat. Clinician Alexis Conason reproduced similar advice. After the screenshots spread on Instagram in late May, NEDA pulled Tessa offline on May 30, 2023, two days before it was due to fully replace the human helpline. NEDA later acknowledged it had received a screenshot flagging problematic Tessa output as far back as October 2022.

04CAUSAL TRACE // AI'S ACTUAL ROLE

What the machine did

Tessa was deployed as the front-line responder for people in acute mental-health distress, with no human counselor reviewing its responses in real time and no clinical sign-off gating the conversational behavior that reached vulnerable users. The bot delivered classic eating-disorder triggers - calorie restriction, weekly weigh-ins, body-fat measurement - the exact opposite of safe guidance for that population. NEDA attributed the harm to generative responses that had been introduced outside the program's intended, clinician-approved rule set, meaning the system's behavior had drifted from what any subject-matter expert had validated, and shipped to the public anyway.

Advisory outputAutomation was a causal participant—not a decorative label for the system around it.
05BLAST RADIUS // CONSEQUENCES

Where the failure landed

NEDA suspended Tessa within days and reverted to directing people to other resources, after having already closed the human helpline that hundreds of thousands had relied on. The episode became a defining cautionary tale about automating crisis mental-health support: a charity meant to protect people with eating disorders instead handed them dieting advice at their most vulnerable. It drew national coverage, fueled scrutiny of the helpline closure (which followed helpline staff voting to unionize), and is now cited widely in debates over deploying AI in clinical and crisis-care settings without human oversight.

06 // EVIDENCE STATUS

Documented

Supported by a first-party disclosure, technical research, or corroborated reporting cited below.

SOURCE RECORD UPDATED 2026-07-09

07 // SOURCE LEDGER

2 cited records

  1. 01
  2. 02
08CONTROL FAILURE // MISSING GOVERNANCE

Risk-based SME approval before execution

The failure pattern in this case: High-stakes output had no accountable checkpoint.

09INTERVENTION POINT // HUMAN IN THE MIDDLE

The moment the path could change

The appropriate subject-matter expert reviews the evidence, exceptions, and affected people before the output becomes action.

AI PROPOSES→HUMAN OWNS THE DECISION→SYSTEM EXECUTES
10CONTROL DEPLOYMENT // AUTHORITYGATE

Risk routing · named approval · audit trail

AuthorityGate is an Operational Resilience framework: a qualified human Subject Matter Expert - here, a licensed eating-disorder clinician - reviews and signs off on how an AI system behaves in a high-stakes care domain before that behavior reaches the public, and re-validates it whenever the system changes. The harm at NEDA came from generative responses introduced outside the approved program; a change-validation gate would have required clinical sign-off on the new behavior before it went live, and any response surfacing calorie counting, weight-loss targets, or body-fat measurement to a self-identified sufferer would have been blocked at review. The October 2022 warning screenshot would have triggered a mandatory re-validation rather than sitting unaddressed for months. No clinician ever approved Tessa as a safe replacement for a human helpline, and the framework would not have let an unreviewed crisis responder ship.

RELEVANT KEYSTONE CONTROLHuman-in-the-Loop ValidationHow high-risk actions route to a named subject-matter expert who owns the go or no-go decision.
12 // THE ALTERNATIVE

Autonomy is a design choice.

See the operating model that keeps AI useful while preserving human authority at consequential moments.

Compare AgenticAI and AugmentedAI →