BRUSSELS, July 7, Nine frontier AI developers were graded on safety by the Future of Life Institute today, and none scored above a C+. Anthropic finished first with that C+, leading five of the six assessed domains. OpenAI and Google DeepMind landed at C. Meta pulled a D+. xAI, DeepSeek and Mistral failed outright.

The Summer 2026 AI Safety Index scored each lab across 37 indicators grouped into six domains: risk assessment, current harms, safety frameworks, existential safety, governance and transparency. Seven independent researchers, including Berkeley’s Stuart Russell and Montreal’s David Krueger, sat on the review panel. Evidence collection closed June 3.

The more revealing finding is directional. OpenAI slipped from a C+ in the previous edition to a C. Anthropic, OpenAI, Google DeepMind and Meta have each weakened or voided earlier commitments to pause development if systems crossed specified danger thresholds. The panel called this pattern “moving goalposts” that has “undermined safety frameworks across the board.” Every major lab was rated “entirely inadequate” on existential risk.

Military work is the other axis of retreat. Companies that banned defense applications before 2024 have quietly reversed course, and Anthropic itself drew reviewer criticism for “questionable military engagements.” The redline that was supposed to be structural turned out to be a phase.

Only five of the nine companies bothered to complete the institute’s survey. Alibaba, xAI, DeepSeek and Mistral didn’t respond.

“AI companies are sprinting toward a cliff,” said Max Tegmark, the institute’s chair.

The next index arrives in six months. The pledges being softened now were, until recently, the industry’s answer to the question of whether self-governance could hold. That answer is being written in the report card.

Sources