Graded on Safety
No U.S. AI company scores above a C+
Independent experts graded the biggest AI companies on safety. Here's how they did.
7 independent experts · 37 indicators · Future of Life Institute · Summer 2026
Share the report card
Post it, forward it, or text it to your group chat.
Copy this caption
Independent experts graded the biggest AI companies on safety four times in under two years. The best score has never gone above a C+. These are the companies racing to build ever more powerful AI – already reshaping work, schools, families, elections, and war. The people building it admit they can't fully control it. It's time to hold these companies accountable. There's a growing movement called Humans in Control – learn more and add your name: https://humansincontrol.org/report-card
Who did the grading
The AI companies say they've got safety handled. Independent experts disagree.
A couple times a year, the Future of Life Institute – an outside research group – has a panel of experts grade the biggest AI companies. And it grades them on the questions that actually matter:
- Do they test for the most dangerous uses before release (like if a model could help someone build a weapon or break into critical systems)?
- Do they have a real plan to keep more powerful AI under human control?
- Does anyone outside the company get to verify their safety claims?
It's not a bad week. It's the pattern.
This week, the top company earned a C+. Not one earned an A in a single category. And this isn't a one-time dip – experts have run these grades four times in under two years, and the best score has never cracked a B.
Someone pays for the safety gap
When these products fail, the company doesn't pay the price. A kid does. A grandparent loses their savings to a scam. A worker trains the AI built to replace them. And as these products get more powerful, the risks get more serious.
A quiet report card doesn't change that. A loud one helps. When enough people share it, it gets harder for the politicians who set safety standards to keep saying the companies have it handled.
Turn a quiet report card loud
Lawmakers act on what voters notice. Every share shows them the public is paying attention – and that pressure is what forces real safety standards.
Humans in Control is building a nonpartisan movement for commonsense rules on AI. Add your name →
