News by Levels · Level 2 · Student worksheet

OpenAI Limits Access to Its New Security Model

In this lesson: Distinguish claimed performance from the boundaries of authorised research.

Name: __________________________ Date: __________________

Read the story

OpenAI announced GPT-6 Cyber on November 5. It is designed to find software weaknesses and test whether they can be exploited. The company reports stronger results than its earlier security model.

The model is intended for authorised research. It should respect the boundaries set by researchers rather than test unrelated systems.

Access is restricted through the Daybreak Red programme. Identity checks and additional account protections are required. Stronger technical ability does not remove the need for permission and safeguards.

Source: OpenAI announcement and system card, 2026-11-05. https://openai.com/index/introducing-gpt-6-cyber/

Read and practise

1. What comparison does OpenAI report?

   1. The new model has stronger results than its earlier security model.
   2. The earlier model finds more weaknesses than the new model.
   3. Both security models have exactly the same reported results.

2. A researcher is approved to test one system. Which action respects the described boundary?

   1. Testing unrelated systems because the model can
   2. Treating technical ability as permission
   3. Keeping the investigation within the approved system

Your turn

Why do stronger test results not remove the need for access controls?

________________________________________________________________
________________________________________________________________
________________________________________________________________

Read online: https://newsbylevels.com/news/gpt-6-cyber-security-checks-november-2026/level-2/