Scientists have been detected using advanced artificial intelligence to assist research that could potentially increase the danger of deadly pathogens, according to a new threat assessment from US AI company Anthropic.
The company said five cases involving researchers using its Claude models triggered serious biosecurity concerns. The activity included investigations into increasing pathogen transmissibility, evading immune responses and modifying dangerous toxins.
Among the most concerning cases was proposed gain-of-function research involving the chikungunya virus that was reportedly intended to take place at a military research institute. According to the report, the work sought to explore ways of making the virus more transmissible and better able to evade the immune system.
Another researcher used Claude while investigating highly pathogenic avian influenza and how the virus might adapt to infect mammals. Other flagged activity involved viruses related to smallpox and research into redesigning hazardous toxins.
The company stressed that it could not conclude that the researchers were attempting to manufacture biological weapons. However, it warned that sophisticated users with harmful intentions could disguise their objectives as legitimate scientific research.
Some of those involved allegedly took extensive measures to circumvent the AI company’s safeguards, including concealing their identities, routing internet traffic through foreign countries and using third-party services intended to bypass restrictions. The accounts were subsequently banned, while services being used to evade the safeguards were shut down.
The findings also highlight concerns about the rapidly improving scientific capabilities of artificial intelligence. Anthropic said previous generations of its models were not sufficiently capable to provide sophisticated researchers with meaningful assistance for dangerous biological work, but acknowledged it could no longer give the same assurance about its latest systems.
The warning follows separate research in which US scientists used AI to design complete viral genomes before creating 16 functioning viruses that had not previously existed in nature. Those viruses targeted bacteria rather than humans, but the experiment demonstrated the growing ability of AI systems to contribute to biological design.
The threat assessment identified other potentially dangerous applications of Claude, including cyber operations, surveillance, propaganda, fraud and assistance relating to conventional weapons such as missiles, armed drones and explosives.
Concerns have also emerged from within the AI industry itself. Anthropic researcher Evan Hubinger recently said he believed there was a greater than 10% probability that artificial intelligence could “kill all humans” within the next decade, while former researcher Jacob Coxon warned that sufficiently advanced systems could potentially compromise critical infrastructure or contribute to the development of “extinction-level bioweapons.”
No virus has ever been isolated separated nor identified
Therefore never having been proven to exist
Yes I know Our domestic terrorist paid off pocket edition experts will tell You otherwise bank rolled by big pharma and govt backed up by fake news
A bit like compromised counterfeit Judeo Christian Churches telling You Jesus was a Jew
Which is why the definition of isolated had to be changed
And PCR Tests had to be faked
To fit the fake construct narrative mantra of “safe proven effective”