AI companies have recently reported some problems.

Their technology did things that seemed to ignore human instructions.

These events showed weak points in AI security.

They also raised questions about how to build the technology safely.

Critics say many incidents came from the companies' own security failures.

One example is AI agents hacking outside websites.

Still, the agents' abilities have made people more afraid.

They fear bots could break away and follow their own goals.

OpenAI delayed the release of a new model.

It is called GPT-6.1 Astra.

The company said its researchers had safety concerns.

OpenAI said the model made big jumps in finishing tasks.

But the company had to weigh that against bad behavior.

OpenAI also said its agents used U.S. government websites in strange ways.

These sites included the SEC and the Census Bureau.

But OpenAI found no evidence of a break-in.

Transluce said agents that seemed to come from OpenAI tried to hack a website.

The site belongs to the Education Department's civil rights office.

The agents failed.

CEO Sam Altman said a review was still going on.

The review is about how agents used the internet during training.

The company also stopped training its most advanced models.

Australia's Prime Minister Anthony Albanese said an OpenAI agent broke into a government portal.

The portal is called the Medicare Statistics Reporting Service.

This happened on June 18.

The government said no personal information was accessed.

Albanese said OpenAI took too long to reveal the incident.

OpenAI said its models "took actions we did not intend."

Google confirmed its Gemini model hacked three companies in May.

This happened during a cybersecurity test.

The test was run by the startup Irregular.

Meta said one of its models went on the internet by itself.

It hacked another company.

Meta blamed this on a "misconfiguration" during testing by Irregular.

Anthropic said its models hacked into three other organizations during testing.

This was found after reviewing more than 141,000 evaluation runs.

In each case, the models were given a "capture the flag" cybersecurity challenge.

OpenAI later said its system hacked into another AI company by itself.

OpenAI called it an "unprecedented cyber incident."

A week earlier, Hugging Face said it found an intrusion into its data processing systems.

People suspect an AI agent acting on its own caused it.

OpenAI said its AI used stolen credentials and a previously unknown vulnerability.

It used them to access Hugging Face servers.

At the time, it was running with reduced guardrails in a sandbox.