Global AI industry falls short on safety, think tank warns

A new report from the Future of Life Institute, a U.S.-based AI safety think tank, highlights significant concerns about the global artificial intelligence industry’s ability to address critical safety issues. The report, published on Tuesday, evaluates the safety practices of nine major AI companies and finds that none meet high standards in addressing potential existential threats posed by advanced AI systems. While some companies showed progress, the findings suggest that the industry as a whole remains far behind what is needed to ensure safe development and deployment of artificial intelligence technologies. According to the report, the evaluation process involved seven researchers and governance experts who assessed the companies based on six categories: risk assessment, current harms, safety frameworks, existential safety, governance and accountability, and information sharing. Each company was rated on these criteria using publicly available data and responses to the think tank’s inquiries. The results show that no company earned an “A” rating in any category, with Anthropic receiving the highest overall score of “C+.” This indicates that even the top-performing entity still has substantial room for improvement in ensuring AI safety. Among the companies reviewed, Meta improved its position by two spots to rank fourth, while xAI fell three positions to seventh place. Other notable placements include China’s DeepSeek in fifth place and France’s Mistral, which finished last. Mistral, a relatively new entrant to the list, expressed disagreement with the methodology used in the report, stating that its approach to developing open-source AI models does not align well with the evaluation framework. Unlike many of its competitors, Mistral allows users to download and modify its AI models freely, whereas companies like Anthropic, OpenAI, and Google DeepMind develop closed models that are not accessible to the public. Max Tegmark, an MIT professor and president of the Future of Life Institute, expressed disappointment with Mistral’s low ranking, noting that European countries have been leaders in AI safety. He emphasized that the institute had made multiple attempts to contact Mistral for feedback but received no response. Similarly, Alibaba, xAI, and DeepSeek were also unreachable during the survey period. Despite this, the report acknowledged that some companies, including three Chinese developers—DeepSeek, Alibaba Cloud, and Z.ai—have produced open models and still ended up in the lower half of the rankings. The report also raised concerns about the reversal of policies by certain companies regarding the use of AI in military applications. Anthropic, for example, was criticized for engaging in questionable military collaborations despite having previously banned such uses. Reports indicate that the U.S. government utilized Anthropic’s technology in military operations in Venezuela and Iran over the past year. However, the Pentagon recently imposed a ban on Anthropic due to disagreements over AI safety protocols. The restriction was temporarily lifted by the Trump administration in late June following pressure from various stakeholders. The Future of Life Institute highlighted that all nine companies evaluated failed to adequately address existential risks associated with the pursuit of artificial general intelligence (AGI)—a level of AI capable of performing any intellectual task that a human can. Although there are some constructive efforts within the industry, the report concluded that these initiatives remain entirely insufficient to mitigate the potential dangers posed by highly advanced AI systems. Additional risks identified in the report include the possibility of AI being misused for cyberattacks or other activities that could harm individuals. These concerns have become more pressing as companies continue to push the boundaries of AI capabilities. For instance, Anthropic recently unveiled its most powerful model yet, named Mythos, which was initially restricted to a select group of trusted organizations due to its potential to expose cybersecurity vulnerabilities. However, the U.S. government intervened in June, blocking the release of Mythos to foreign entities on national security grounds. The ban was later lifted by the Trump administration, allowing Anthropic to proceed with further development and distribution of the model. As the debate around AI safety continues to gain momentum, the report underscores the urgent need for stronger collaboration among governments, private sector entities, and research institutions to establish robust safeguards against emerging threats. With the rapid pace of technological advancement, the global community must act decisively to ensure that AI developments benefit society without compromising safety and ethical considerations.