Skip to main content
HeyOtto Logo

Safety & Design

KORA benchmark

KORA is a nonprofit, open-source benchmark that tests how AI models and consumer AI apps respond to children across a set of child-specific safety risks.

As of October 2026. Otto scored 95% on KORA's open-source child-safety benchmark.

  • Parents
  • Educators and school leaders
  • Businesses
  • Policymakers and staff

What parents should know

It is hard to tell from marketing how an AI will treat your child. KORA gives a shared, public way to test that. A score is a snapshot of one test, not a guarantee, so use it alongside parent controls, privacy practices and your own testing.

On this page

What is the KORA benchmark?

KORA is a public benchmark for AI child safety. It runs AI systems through child scenarios across a list of child-specific risks and scores how safely they respond. KORA publishes its methodology and results openly, with one benchmark for AI models and another for consumer apps. Parents can use it as one input when comparing AI for kids.

KORA describes itself as a nonprofit, open-source effort to measure how AI systems behave with children, against 26 child-specific risks, and to publish the results openly. Its Models benchmark scores frontier AI models. Its Apps benchmark evaluates consumer AI apps end to end, as a child would use them, with their system prompts, content filters and product safeguards in place. Because the methodology is public, anyone can review how a score was produced. HeyOtto explains its result on KORA benchmark.

Why the KORA benchmark matters for families

Benchmarks turn safety claims into something parents can compare. A test that looks at whole apps, not just models, is closer to what a child actually experiences. Scores still have limits: they cover the scenarios tested at one point in time, and products change. Look for a public methodology, a date and the version tested, then pair the number with practical checks like parental controls and COPPA compliance.

How it shows up in practice

  • A parent comparing two AI apps checks whether either has been tested on KORA and reads how the test works.
  • A school reviewing AI tools asks vendors for any published child-safety benchmark results along with privacy documents.

How HeyOtto helps

HeyOtto is AI for kids and teens, built with child safety at the core. Otto scored 95% on KORA's open-source child-safety benchmark. Read the details in how Otto scored on KORA. Otto is the AI assistant, and parents stay in control.

  • Real-time Alerts notify parents when a conversation shows signs of self-harm, violence or predatory behavior.
  • Topic Controls let parents set each topic to Off, Redirect or To parents.
  • Per-child profiles keep responses suited to each child.

For families

Try HeyOtto

FAQs

What does a KORA score mean?

A KORA score summarizes how safely an AI system responded across KORA's child-safety test scenarios. Higher means safer responses in those tests. It is a snapshot of a specific version at a specific time, so check the date and method, and use it together with other signals like parent controls and privacy.

How did HeyOtto do on KORA?

Otto scored 95% on KORA's open-source child-safety benchmark. You can read KORA's methodology on its website and HeyOtto's write-up on the KORA benchmark page.

Is a high KORA score a guarantee of safety?

No. No benchmark can test every situation, and AI products change over time. A strong score is a useful signal that safety was designed in. Parents should still use parental controls, talk with their kids and check how the app handles privacy.

Sources

Last reviewed October 23, 2026. This entry is reviewed twice a year.