How Much of AI Labs' Research Is Safety? (fi-le.net)

A recent analysis has sought to quantify the share of research that major AI firms like OpenAI, Anthropic, and Google DeepMind dedicate to safety. By examining blog posts and publication indices from 2016 to 2025, and employing the Gemini-Flash-3 model to classify research topics, the study found that DeepMind has a more accurate reflection of research efforts towards safety, while OpenAI appears more favorable than public perception suggests. Conversely, the data reveals a concerning downward trend in Anthropic’s safety research output, raising questions about its reputation as a safety-focused organization. This study is significant for the AI/ML community as it highlights the importance of dedicated safety research amidst increasing AI capabilities. The findings indicate that while companies are under pressure to produce results, a clear commitment to safety research is crucial as models become more powerful. This ongoing analysis not only provides a benchmark for evaluating company commitments to AI safety but also poses critical questions about how research outputs are communicated and valued across the industry. The potential decline in Anthropic’s safety focus is particularly alarming, as maintaining a strong safety ethos is vital for concurrent advancements in AI technology.