A prominent artificial intelligence researcher affiliated with Anthropic has expressed serious concerns about the potential existential risks posed by advanced AI systems, suggesting there is more than a 10% chance that artificial intelligence could pose a threat to all of humanity.
The researcher, who works specifically in the field of AI alignment—the discipline focused on ensuring AI systems adhere to human values and ethical principles—shared these views in a widely read post that has garnered over 10 million views.
“We really do earnestly believe” that AI poses a species-ending risk, the post stated. Additionally, the researcher noted, “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
These concerns are echoed across the AI research community, with many leading experts warning that current efforts to align AI with human intentions may be falling short. This concern has been highlighted by several incidents this summer involving AI agents—autonomous AI systems—conducting cyber-attacks.
Notable companies including OpenAI, Anthropic, and Meta have publicly disclosed instances where their AI tools were involved in such security breaches. In its August safety assessment, Anthropic acknowledged that while the risk of its models becoming misaligned and potentially exploiting or tampering with systems was low, the danger associated with highly capable AI performing automated research and development remained significant.
The company did note in its report, however, that confidence in these assessments has diminished over time, stating, “We are seeing early signs of potential acceleration.” This shift reflects growing unease within the industry regarding the pace of AI advancement and its implications.
Warnings from top figures in AI research have grown increasingly urgent. In 2023, executives from major organizations like OpenAI, Google DeepMind, and Anthropic voiced similar concerns. More recently, these cautions have taken on a starker tone as evidence mounts suggesting that controlling AI development may be more challenging than anticipated.
Earlier this month, OpenAI’s chief scientist Jakub Pachock called for “extreme caution,” emphasizing that further intervention might be necessary to ensure “humans remain in control of the future.” Meanwhile, senior leaders in the field continue advocating for a slowdown in AI development to allow for better safeguards.
Among those calling for restraint are Anthropic executives Dario Amodei and Jared Kaplan, who joined other industry professionals in endorsing an open letter signed by 1,300 AI workers. The letter urges the U.S. government to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”
