If you are a business leader evaluating which AI vendors to trust with your operations, customer data, or internal workflows, there is a report worth adding to your reading list. The Future of Life Institute released its Summer 2026 AI Safety Index in early July, grading nine of the world’s leading AI labs on how seriously they take safety.

The highest grade went to Anthropic. They received a C+.

What the Index Found

The Future of Life Institute assessed nine AI developers across six domains: risk assessment, current harms, safety frameworks, existential safety, governance, and information sharing. Grades use the standard US GPA scale, from A+ down to F.

Here is where the major companies landed:

• Anthropic: C+ (highest overall, leads five of six domains)
• OpenAI: C (slipped from C+ compared to the previous edition)
• Google DeepMind: C
• Meta: Improved from 6th to 4th place
• xAI, DeepSeek, Mistral: Failing grades

Anthropic’s edge comes from stronger transparency, a more established safety framework, technical research output, and governance structures. OpenAI leads specifically in risk assessment, which the index credits to a broader evaluation suite and diverse external testing engagement.

Three companies received outright failing grades. One is from the US (xAI), one from China (DeepSeek), and one from Europe (Mistral).

The Finding That Should Concern Enterprise Buyers

The grades themselves are notable, but the trend the index identified underneath them is more significant for anyone making procurement decisions.

From 2024 to 2026, companies including Anthropic, OpenAI, Google DeepMind, and Meta — all of which previously maintained policies banning military applications — gradually reversed course. All four are now actively pursuing defense partnerships, joining xAI and Mistral, which never held those restrictions.

The practical meaning for business: the companies setting AI safety standards are also the companies under the most commercial pressure to loosen those standards. The index’s researchers summarized this bluntly. The voluntary safety system created by AI labs has begun eroding before governments have put a durable alternative in place.

Existential safety is the weakest domain across the entire industry. No company scores higher than C- in this category. Most score D or below.

What This Means for Business

This report should not be read as a reason to avoid AI. That conclusion misses the point. The right reading is more nuanced and more useful.

Vendor selection matters more than category selection. The gap between a C+ and a failing grade is not trivial. When you are deploying AI agents that touch customer interactions, financial data, or internal operations, the safety culture of the model provider is a real variable — not a marketing consideration.

Compliance requirements are coming. The AI Safety Index exists because governments and regulators are looking for anchors. The EU AI Act, US executive orders, and emerging industry standards are all moving toward formal accountability requirements. Companies that are already ahead of this curve — however modestly — are better positioned when those requirements harden.

Transparency is a useful signal. The index rewards companies that share more about how their models work, what they test for, and how they evaluate risk. When evaluating AI tools, the same logic applies: vendors willing to answer hard questions about safety and data handling are different from vendors who can’t or won’t.

The best-in-class is still a C+. This is the finding that should prompt the most reflection. A C+ is not a passing grade in most academic contexts. It means there is real and meaningful risk in the tools your business is adopting, even when you choose well. That is not an argument for paralysis. It is an argument for building internal processes that do not rely entirely on AI doing the right thing. Human review, audit trails, and clear escalation paths are not optional extras in this environment.

The Bigger Picture

The AI Safety Index is published twice a year. The Summer 2025 edition represented a baseline. The Summer 2026 edition shows movement, but not in a uniformly encouraging direction. The companies that are improving are doing so modestly. The companies that are sliding are doing so in ways that correlate with competitive and commercial pressure.

For business leaders, the takeaway is not which AI company to root for. The takeaway is that AI vendor evaluation is now a legitimate part of risk management, and the infrastructure to do it is being built in real time.

The Future of Life Institute is doing some of that building. The grades are rough. The methodology is imperfect. But the project of holding AI labs publicly accountable for their safety practices is more valuable than its current report card reflects.

---

Enterprise DNA helps businesses evaluate, implement, and govern AI responsibly. If you are thinking through your AI vendor strategy or governance framework, start a conversation with our advisory team.