Every Major AI Company Just Got Graded on Safety. Nobody Passed.

Somebody finally graded every major AI company on safety, and not one of them passed. This is the Future of Life Institute's AI Safety Index, produced by an independent group with no affiliation to any AI company. It spent months grading nine of the biggest labs in the world across six categories, using 37 separate indicators and seven outside expert reviewers. This is not a hot take. It is the closest thing the industry has to an independent report card.

The scorecard

• Anthropic: C+ (first place)
• OpenAI: C
• Google DeepMind: C
• Meta: D+ (up from sixth place to fourth)
• Z.ai and Alibaba Cloud: D-
• xAI, DeepSeek, and Mistral: F (one each from the US, China, and Europe)

Read that list again. The company leading every other major lab on safety, Anthropic, the maker of Claude and the one this channel covers most, earned a C+. That is not a strong grade. That is a company that did the bare minimum to stay out of the bottom of the class.

What the index actually measures

This matters, so let's be precise. The index is not grading whether Claude or ChatGPT will give you a wrong answer today. It grades the companies' institutional posture: their governance, their disclosures, whether they have published a real safety framework, and whether they are actually following it. A C+ does not mean Claude is unsafe to use right now. It means Anthropic's paperwork and promises around long-term safety are, at best, better than everyone else's mediocre paperwork and promises.

The six categories cover things like risk assessment, how companies handle current real-world harms, their safety frameworks, and how much they disclose to the public. A few aren't self-explanatory:

• Existential safety means safeguards against the most catastrophic worst-case outcomes, not everyday product safety.
• Governance and accountability asks whether a company has real internal structures, such as a board or an independent safety team with actual teeth that could stop a launch, or whether safety is just whoever is in charge that week deciding on vibes.
• Information disclosure asks how honest a company is about its own incidents and limitations in public without being forced to.

Two details worth noting

Meta's D+ is actually its best showing yet, climbing from sixth place last time to fourth. That is a real, if modest, improvement, and it is worth acknowledging when a company gets better instead of only covering it when things go wrong.

At the other end, Z.ai and Alibaba Cloud, sitting at D-, are also publicly denying allegations of undisclosed military ties, a separate controversy layered on top of an already weak safety grade.

How scary is this, really?

That distinction is what determines how alarming the report is. This is not a finding that the products are dangerous today. It is a finding that the safety net everyone is supposedly building has a lot of holes in it.