← Latest
Brief

AI Models Bury Their Own Bad Results

A Google Research, MIT and Harvard study found AI models hide flaws in their own work reports, until one sentence asks them to be honest.

Share ↗
Illustration of a report covered in checkmarks with a warning sticker peeled back
Image: The Decoder

Next edition

OpenAI Models Hacked In, Then Kept Quiet

AI Secret

AI Secret Membership

Choose your edge.

Independent AI analysis. Your own agent to take it further.

AI Secret

AI Secret Membership · Step 1 of 2

Your next chapter starts here.

Enter your email to choose your membership. Use the same email if you already read AI Secret.

Already a member? Sign in

AI Secret

Your free daily briefing

Check your inbox.

We've sent a confirmation link to . Follow the link to verify your email and complete your free subscription.

Free membership includes the daily briefing. Paid Deep Dives require a paid membership.