Artificial intelligence companies, including Anthropic and OpenAI, are bringing in chemical weapons and explosives experts as they work to limit the risk of harmful outputs from advanced AI models.
The move reflects growing concern that powerful AI systems could inadvertently assist users researching chemical, biological, or radiological weapons.
According to a BBC report, Anthropic advertised a role seeking a specialist in chemical weapons or explosives defense with knowledge of radiological dispersal devices, or so-called “dirty bombs.”
The position aims to strengthen safeguards that prevent its AI systems from generating sensitive or potentially dangerous information.
OpenAI has also put out a similar job posting focused on biological and chemical risks, offering up to $455,000 per year for the role.
The Exposure Dilemma
While companies aim to harden their systems against harmful queries, testing and evaluating models against chemical, biological, or radiological scenarios may require exposing them to the very categories of information they are designed to restrict.
That tension has drawn attention beyond the private sector. Military officials have raised concerns about sensitive weapons-related data entering AI systems.
The US Navy Explosive Ordnance Disposal Technology Center recently cautioned personnel against uploading bomb disposal procedures into generative AI tools.
A separate study by the RAND Corporation has also warned that advances in AI could lower the barrier for malicious actors seeking to plan or carry out chemical or biological attacks.
