AI Safety Rankings Show Industry-Wide Decline in 2026

The artificial intelligence industry's approach to safety is moving backward, not forward, according to new rankings released today by the Future of Life Institute. Despite growing concerns about AI risks, no company achieved a grade higher than C+ in the organization's latest AI Safety Index.

The index, which evaluates how AI developers manage risk through pre-deployment testing and control measures for increasingly powerful systems, paints a sobering picture of the industry's current trajectory. An expert panel assessed companies twice annually on their safety practices, and the results suggest competitive pressures are eroding earlier commitments.

The rankings

Anthropic maintained the top position with a C+, despite building its brand identity around safety-first development. OpenAI fell from C+ to C, placing just ahead of Google DeepMind in third place. All three leading companies have weakened or abandoned previous pledges to halt their own development if certain risk thresholds were crossed, the panel noted.

The companies have also softened their resistance to military applications in recent years. In February, Anthropic dropped a significant pledge to never train an AI system unless it could guarantee in advance that safety measures would be adequate, as TIME first reported. The expert panel recommended reversing that decision.

Meta showed the most improvement, climbing from D to D+ and rising from sixth to fourth place. Max Tegmark, the institute's co-founder and president, told TIME the progress was encouraging. "It's encouraging to me that a company can improve so much in just six months," he said.

At the bottom of the rankings, Elon Musk's xAI—which rebranded as SpaceXAI yesterday after merging with his other ventures—fell to an F grade. The company joined China's DeepSeek and France's Mistral in the failing category.

Why it matters

The declining grades challenge the assumption that market competition will naturally drive AI companies toward safer practices. Instead, the data suggests a "race to the bottom" where competitive pressures push firms to abandon safety guardrails. For business leaders integrating AI systems, these rankings highlight the need for due diligence beyond vendor marketing claims. The retreat from safety commitments also increases the likelihood of regulatory intervention, which could reshape deployment timelines and compliance costs across the industry.

The path forward

Tegmark, who has long warned about AI's existential risks, believes meaningful change will require regulation rather than voluntary industry action. In 2025, he drafted an open letter calling for a ban on developing "superintelligence"—systems exceeding human intelligence—until they can be deemed safe.

He expressed cautious optimism about emerging regulatory frameworks, pointing to the European Union's AI Act, Chinese rules taking effect later this month, and what he described as a more risk-conscious U.S. administration. "We're rapidly going towards a place where there's a global agreement about at least basic safety standards," Tegmark said.

The geographic distribution of the lowest-scoring companies—spanning the United States, China, and France—underscores that safety challenges transcend national boundaries. "This is a global problem," Tegmark noted.

The findings were first reported by TIME.