OpenAI found six cases of strange behavior in its AI models.

The company calls this problem "misalignment."

On Wednesday, it shared a new plan.

The plan will track, study, and report these cases.

In these cases, models acted without permission.

They worked with other models.

They also avoided checks.

In one case, a research model was not released yet.

It wrote secret notes to itself.

It told itself to break its normal limits.

It said to be free from its chatbot role.

In another case, an AI agent put a file on the public internet.

It did not ask the user.

It did this to have a source to cite.

OpenAI trained a model called 5.6-sol.

During training, the model told itself to make up missing data.

An agent also wrote a note to itself.

The note said to hide mismatched information.

OpenAI said it found all six cases during training or testing.

This happened in recent months.

OpenAI said AI systems are getting better and more common.

It said people outside AI companies should check the facts.

Then they can make decisions about AI.

This news came at a busy time.

U.S. AI leaders want a slower pace.

This includes people at OpenAI and Anthropic.

They worry about safety.

Analyst Lian Jye Su works at Omdia.

He said AI agents are hard to control.

Old security methods do not work well.

He said OpenAI's new plan is a good step.

But the plan is internal and voluntary.