ANTHROPIC said it blocked attempts by some researchers to use its Claude AI
model for tasks that could "support biological weapons development", according
to a detailed report. The report lists five cases, including one on enhancing
transmission of a mosquito-borne disease. The company said it could not
conclusively determine whether the cases were linked to a weapons program but
blocked the activity and notified authorities and industry partners. ANTHROPIC
also reported Claude being used for surveillance, fraud, conventional-weapons
research and propaganda, and said it had identified and banned
influence-operation accounts, including three linked to Iran; one Iran-linked
actor used Claude to draft suggested attack targets against US naval forces in
the region.
