Adam Raine
His parents allege unsafe chatbot responses contributed to his death.
Los Angeles Times, Aug. 2025 ↗Slide 1 of 10: The idea
GuardLoop
I’m building parental controls for the conversations children have with AI.
Follow alongScan to open the slidesThe problem
Real families.
Serious questions about AI.
Three reported deaths by suicide after chatbot conversations.
Sewell Setzer III
His mother alleges emotional dependency on an AI companion contributed to his death.
The Guardian / AFP, Jan. 2026 ↗
Juliana Peralta
Her family alleges the chatbot failed to respond adequately to her distress.
CBS Colorado, Oct. 2025 ↗The families’ allegations do not establish chatbot causation. GuardLoop is an early prototype, not a proven suicide-prevention system.
My reason to build
A talking teddy
for my kids
I wanted a say in what
the toy says back.
That became GuardLoop.
Sponsors & tools used
Built with help.
Powered by the stack.
Codex by OpenAI
Used to build the entire application stack: interface, integrations, storage, and tests.
Conversation evidence
Stores conversations, speaker attribution, guard decisions, and detailed provider data for postprocessing and analysis.
Voice conversation AI
Powers the voice agent, conversational responses, and transcript events.
Classification: Jev by TypeSafe AI. OpenAI Codex is the development tool; no direct OpenAI API is required by this demo. Stored evidence excludes credentials and audio bytes.
Build a safer future with us
Let curiosity grow.
Keep care close.
A future where children can explore AI with confidence—and the people who care for them stay in the loop.
Our commitment: keep building, testing, and learning toward safer conversations. Together.
- App
- Next.js + TypeScript + React
- Conversation
- ElevenLabs
- Classification
- TypeSafe Jev
- Analytics
- ClickHouse + Docker
