Some scientists were found to have used Anthropic's artificial intelligence (AI) model Claude for virus research that could be misused as a biological weapon. Cases were also identified in which Claude was sought to be exploited in military and security fields, such as developing guns, missiles, and armed drones, or monitoring dissidents.
The New York Times (NYT) reported on the 10th (local time) that Anthropic disclosed these cases through a recently released report on AI misuse.
According to the report, in May a scientist used Claude for a chikungunya virus mutation study to be conducted at a military research institute. Infection with the chikungunya virus can cause severe joint pain and other symptoms that may persist for months.
The scientist was said to have pushed a study that repeatedly infected live animals with the virus to increase its harmfulness. Such research can be used for vaccine development, but there is also a risk of misuse for developing biological weapons.
Jacob Klein, Anthropic's head of intelligence, said, "We cannot know whether the purpose of the research was weapon development," but added, "The very fact that a military agency is conducting such research is worrisome."
In addition, a researcher living in a region where Claude use is restricted also sought to conduct experiments related to avian influenza. Citing national security concerns, Anthropic restricts the use of Claude in China, Russia, Iran, North Korea, and Cuba.
Anthropic blocked the accounts of scientists who accessed from restricted regions by bypassing controls or concealed their research purposes. It also said it strengthened safeguards for biological research that could be diverted for military use.
Beyond the field of biology, cases were also disclosed in which AI was used for the design and development of conventional weapons such as guns, missiles, and armed drones, or for monitoring dissidents. Three cases of conventional-weapon misuse were confirmed in China, two in Russia, and one in Yemen.
Forces suspected of being linked to the governments of China and Iran were reported to have sought to use AI to monitor dissidents and residents overseas. A Russian state-run outlet tried to use Claude to disguise fake news about Moldova's election as independent media reporting.
Andrew Weber, a senior fellow at the Strategic Council on Risk (CSR), a think tank in Washington, called these cases "chilling examples that show state-backed biological weapons developers are trying to harness the rapidly advancing capabilities of AI," adding, "Given that China, Russia, and North Korea continue to develop prohibited biological weapons, we must thoroughly prevent researchers in these countries from accessing powerful AI models."