Anthropic is hiring a weapons expert to decide what Claude should be allowed to do

Anthropic is hiring a weapons expert to help decide what its Claude AI models should and should not be allowed to do.

The unusual job, called Policy Design Manager, Conventional Weapons, sits within Anthropic’s Safeguards organisation. The successful candidate will help define the boundary between legitimate engineering research and requests that could contribute to weapons development. The position comes as Anthropic faces growing questions about how increasingly capable AI systems could be misused in military and other high-risk settings. And the company is willing to pay heavily for someone who can help answer those questions.

Anthropic is offering up to $285,000

The job listing gives an annual salary range of $245,000 to $285,000. At the upper end, that is roughly Rs 2.7 crore a year, depending on the exchange rate. That works out to about Rs 22.5 lakh a month before tax. Anthropic says it is looking for someone with deep, practical knowledge of weapons systems. Experience at a defence research laboratory, government research agency or weapons company is among the backgrounds it lists. A bachelor's degree or equivalent experience is required. Knowledge of machine learning is preferred, along with experience in areas such as robotics, autonomy, guidance and navigation, sensors, aerospace or embedded software.

What will the weapons expert actually do?

The job is not simply about blocking questions about guns or military equipment. Anthropic says conventional weapons increasingly depend on software and autonomous systems. The role therefore covers technologies with legitimate civilian uses that could also contribute to weapons systems. The person hired will develop threat models and evaluations to assess how Claude could contribute to weapons development. They will also work with engineers to turn policy decisions into model guardrails, detection systems and enforcement tools.

Anthropic specifically says the role extends to autonomous systems capable of selecting and engaging targets without human authorisation.

The aim is to establish a practical line between acceptable research and harmful weapons development.

Why is Anthropic doing this now?

The timing is significant. Anthropic has been expanding its AI safeguards as its models become more capable. Last week, the company published research showing that frontier AI models can perform tasks relevant to tactical intelligence and conventional weapons development, while also describing safeguards designed to block misuse. Anthropic has also previously restricted Claude from assisting with weapons design and development. The company has argued that advanced AI systems need stronger protections in high-risk areas.

That makes the new hire part of a broader effort to build safeguards around Claude rather than evidence that Anthropic believes an AI war is about to begin. The bigger question is becoming harder to ignore: as AI becomes capable of handling more sophisticated technical work, where exactly should the line be drawn? Anthropic is now looking for a weapons expert to help draw it.