AI Companies Quietly Drop Safety Promises While Building More Powerful Models, New Report Warns

A new report released today has raised fresh concerns about the direction of the artificial intelligence industry, warning that some of the world's leading AI developers are quietly moving away from safety commitments they previously promoted. 

The report, published by the Future of Life Institute, argues that the rapid race to build increasingly powerful AI systems is now moving faster than efforts to ensure those systems remain safe, transparent, and accountable.

The study evaluated major AI companies including OpenAI, Anthropic, Google DeepMind, Meta, xAI, and several other developers using dozens of indicators covering governance, transparency, cybersecurity, model evaluations, and long-term AI risk management. Researchers concluded that while technical capabilities continue improving at an extraordinary pace, progress on safety has not kept up with the speed of development.

One of the report's most significant findings is that several companies have weakened or removed voluntary commitments they previously made regarding frontier AI development. Earlier public pledges suggested that companies would slow or pause development if future AI systems reached certain danger thresholds. According to the report, many of those commitments have now been revised, reduced, or abandoned as competition within the industry intensifies.

Anthropic received the highest overall score in the report but still achieved only a C+ rating, while OpenAI and Google DeepMind both received grades in the C range. Researchers said no company demonstrated comprehensive safeguards across every category that was evaluated, highlighting how much work remains despite billions of dollars being invested in AI development.

The weakest area across nearly every company involved what researchers described as existential safety—preparations for extremely advanced AI systems that could create severe societal or security risks if they behaved unpredictably or were misused. The report argues that companies remain heavily focused on releasing increasingly capable models while investing comparatively less attention in long-term safety planning.

The findings arrive at a time when governments around the world are debating how artificial intelligence should be regulated. Policymakers in the United States, Europe, the United Kingdom, and Asia are developing new AI laws covering transparency, copyright, cybersecurity, national security, and consumer protection. Despite these efforts, researchers say regulatory frameworks still lag behind the pace of technological innovation.

Another issue highlighted in the report is the growing involvement of AI companies in military and defense projects. Several firms that previously expressed caution about military applications are now expanding partnerships involving national security, autonomous systems, cybersecurity, and intelligence analysis. Researchers believe these developments further increase the importance of establishing clear international standards for responsible AI deployment.

Industry leaders continue to argue that artificial intelligence has enormous potential to improve healthcare, education, scientific discovery, finance, software engineering, and productivity. However, experts increasingly warn that rapid commercial competition should not come at the expense of robust testing, transparency, and independent oversight. As companies compete to release more capable AI models, balancing innovation with public safety is becoming one of the technology industry's biggest challenges.

The report is expected to fuel further debate among regulators, investors, researchers, and technology executives over whether voluntary commitments are sufficient or whether stronger legal requirements are now necessary. With increasingly powerful AI systems expected to launch over the coming months, the conversation around AI safety is likely to become just as important as the race to build the next breakthrough model.