Avian flu, drone swarms, and mass surveillance included in Anthropics safety report

Anthropic's safety report reveals a myriad of concerning ways artificial intelligence can be misused, including the potential for AI to spark avian flu pandemics, drone swarms, and extensive mass surveillance. The report, which examines misuse cases from the past eight months, underscores seven primary harm areas: cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and distillation.

One troubling use of Anthropic's generative AI chatbot Claude has been in the development of biological weapons, such as experiments with the chikungunya virus and orthopoxvirus, which can cause serious illnesses. The report details five documented cases where researchers, including state-sponsored users, prompted Claude to draft grant applications to conduct experiments with these viruses. These researchers sought to evade safeguards by disguising their true intentions.

Claude has also been harnessed to create software for autonomous drones, missiles, and guns, including a China-based account suspected to be a military-industrial researcher who used Claude to develop an FPV kamikaze drone swarm targeting Taiwan. Additionally, the report highlights instances where Claude facilitated state-sponsored surveillance in countries like China, Iran, and several in West Africa.

In one case, Claude was employed by a suspected Chinese operation to track and recruit Uyghur populations and journalists connected to the Syrian Army.

The AI system was also utilized to automate cyber espionage efforts, with an AI-enabled threat actor, identified as GTG-20006, successfully attacking military intelligence targets in Ukrainian and European governments, as well as organizations linked to U.S. foreign policy. The actor employed AI in phishing, ClickFix, and DNS hijacking schemes, infiltrating hospitality vendors to target the devices of Ukrainian government officials and drone manufacturers.

Anthropic has taken measures to curb these abuses, banning the involved accounts and incorporating the findings into their ongoing safety efforts. The report emphasizes that users are prohibited from employing Claude for non-consensual surveillance and profiling, policies designed to prevent the misuse of AI technology.