EVERYTHING
According to Bankinfosecurity, the Future of Life Institute has released its latest AI Safety Index, revealing a disappointing landscape for the world's leading technology firms. The independent panel of seven experts evaluated nine major developers against 37 indicators across six key domains, including risk assessment, existential safety, and governance. The findings suggest that while some companies are slightly ahead of others, none have reached the threshold of responsible practice defined by the reviewers.
Anthropic leads a field of underperforming giants
Despite receiving an overall grade of C+, Anthropic secured the top spot in the index. The report attributes this leadership to the company's robust transparency practices and its established safety framework. Notably, Anthropic is currently the only developer that publicly shares both its system prompt and a behavior specification. The firm also earned a B+ in information sharing, which was identified as the highest individual mark awarded across all companies in the study.
Other major players followed closely but with lower marks: OpenAI took second place with a C grade, while Google DeepMind followed with an identical C rating. Meta showed some improvement, rising from sixth to fourth place with a D+ grade. However, xAI saw a significant decline, dropping to seventh place and receiving an F grade due to what reviewers described as "gaping holes" in its risk-assessment evaluations.
Weakening pledges and competitive pressures
The report highlights a concerning trend of "moving goalposts," where companies are retracting previous commitments to pause development if specific risks are met. The review identified several critical issues with current industry standards:
Safety investigator Sabina Nong noted that if safety remains governed solely by corporate-defined rules without independent oversight, the industry risks prioritizing speed over security. The report underscores that inadequate safety is a global issue, with failing grades distributed across companies based in the United States, China, and Europe. This assessment serves as a stark reminder that current self-regulation may be insufficient to manage the rapid evolution of artificial intelligence.