Skip to content
Submit incident
AII-1000-44047
Under investigationA regulator, court or police has opened a formal inquiry and reached no finding yet.

Discord moderator AI failed to flag server in Brazilian teen's death

A 13-year-old girl in Brazil died in a Discord live broadcast on 22 July 2026, reportedly after being induced to suicide. Discord's moderation AI failed to flag the newly-created server for human review, and the company acknowledged the delay. The Brazilian data authority ordered Discord to suspend live streaming in the country.

22 July 2026 · Discord · Safety and wellbeing
22 July 2026
AII-1000-440472 sources ↓
Quick facts
Category
Safety and wellbeing
Country
Brazil
Source
O TEMPO
Date
22 July 2026
Drafted by
Raidu Incident Detection

What happened

A 13-year-old girl died on 22 July 2026 in Naviraí, Mato Grosso do Sul state during a live broadcast on Discord. According to reports, she was allegedly induced to suicide through content in a Discord server.

Discord's vice-president of Safety and Trust, Jud Hoffman, acknowledged that the server where the broadcast occurred was newly created (less than two hours old) and small, and was not flagged by the AI moderation system for human review. Hoffman stated: 'The information available did not trigger our machine learning model to send it for human review. We were not fast enough, but we are always working to improve those models and security efforts.'

Reported impact

The death of a 13-year-old girl occurred during a live stream on the platform. The Polícia Civil de Mato Grosso do Sul is investigating the circumstances of the death and the actions of members of the virtual group.

Index analysis

Discord acknowledged that its AI moderation failed to identify and escalate the new, small server for human review in time. Hoffman noted that the platform uses both proactive and reactive moderation combining machine learning models with specialized teams, with approximately 15% of the company working on security across hundreds of moderation roles, including Portuguese-speaking staff familiar with the Brazilian market.

Hoffman attributed part of the difficulty to scale and the balance between detecting more risks and avoiding false positives. He stated that initial signals can come from user reports or authorities, and machine learning models help prioritize what goes for human review.

Governance controls

Following the incident, the ANPD (Agência Nacional de Proteção de Dados) ordered Discord to suspend live video streaming on the platform in Brazil. Hoffman stated that the company will comply with the determination and apply the block consistently across all video resources available in the country.

Classification

Category
Safety and wellbeing
Industry
Media, social platforms and marketing
Country
Brazil
Harm status
Actual harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Physical death, Psychological, Human or fundamental rights
Harmed
Children
Failure modes
Content moderation failure
Consequences
System suspension or termination
AI systems
Discord AI moderation system

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

O TEMPO
'Não fomos rápidos o suficiente', admite representante do Discord sobre morte de menina ao vivo
15 August 2026
uol.com.br
'Não fomos rápidos o suficiente', diz Discord sobre adolescente morta
15 August 2026
Dec 2024Texas AG launches investigation into Character.AI and other platforms over child safetyUnder investigationSafety and wellbeingNew York PostThe News InternationalmySASep 2026Delhi man arrested for creating and circulating obscene AI-generated images of womanConfirmedHarmful and abusive contentThe Times of IndiaKalingaTVAsianet News Network Pvt Ltd+2Aug 2026German Intelligence Warns of AI Algorithms Fueling Online RadicalizationConfirmedHarmful and abusive contentT-online.deAug 2026ANPD Orders Suspension of Facial Recognition in Paraná SchoolsConfirmedPrivacy and surveillanceOlhar Digital - O futuro passa primeiro aquiCartaCapitalJul 2026Brazilian Ministry identifies 32 AI sites creating fake nude imagesConfirmedHarmful and abusive contentMarcelo BittencourtBrasil 247Folha de S.Paulo