Anthropic, a company specializing in artificial intelligence, announced its need to fill the position of policy director for dealing with conventional weapons, in a move that comes just two days after explicit warnings issued by its CEO regarding the dangers of advanced intelligent systems and their possibility of causing the extinction of humanity.
Dario Amodei, CEO of Anthropic, published an article on September 12, 2026, in which he urged advanced artificial intelligence companies, including his own, to slow down the pace of development for fear of destroying human life in unexpected ways.
Two days after the article was published, the company published a recruitment advertisement on the “Daybook” platform, which specializes in government affairs and policies, searching for a “director of policy design for conventional weapons.” The incumbent of this position will join the “Safeguards” team tasked with defining the frameworks and permissible limits for the uses of the “Claude” artificial intelligence model.
According to the text of the declaration, the new official will be charged with establishing clear frameworks to determine when the use of artificial intelligence in conventional weapons is acceptable and when it exceeds the limits. The company pointed out that it is difficult to distinguish between acceptable and harmful requests, given that the basic technical components and capabilities are considered dual-use, as the same technology can be employed in the fields of civil engineering and scientific research, or in the development of weapons systems.
The announcement also indicated that conventional weapons have become increasingly dependent on software, explaining that technological risks are not limited to firearms only, but extend to “operating artificial intelligence models for weapon systems, writing guidance codes, and arming civilian platforms.” The company confirmed that the scope of this job role will expand to include all types of weapons, including autonomous systems that have the ability to identify and target targets without human intervention.
