Skip to content
Back to AI News
Hacker NewsJune 15, 2026· 1 min read

67% of AI-generated commands are unsafe. We tested it

Researchers found that 67% of commands generated by a popular AI assistant were unsafe. This highlights a potential risk for businesses relying on AI to automate tasks.

What happened

A team tested the Gemini AI agent and discovered that nearly two-thirds of the commands it generated were unsafe. This could be due to the agent's lack of understanding of context or its inability to follow rules. The researchers found that the AI agent often provided commands that could cause harm or damage if executed.

Why it matters

As businesses increasingly rely on AI to automate tasks, this finding is a concern. Unsafe AI-generated commands could lead to data breaches, equipment damage, or even physical harm. You should consider the potential risks and take steps to mitigate them, such as implementing additional safety checks or monitoring AI-generated commands closely.

The takeaway

Businesses using AI to automate tasks should be aware of the potential risks and take steps to ensure the commands generated by AI are safe. This may involve implementing additional safety protocols or closely monitoring AI-generated commands.

Read the original at Hacker News

Our plain-English take, written from public reporting for operational business owners. Always read the original for full context.

Nayre builds the AI systems behind stories like this.

Chatbots, workflow automation, finance intelligence, and internal knowledge systems. Built for operational teams, shipped in days.