On Thursday, Anthropic published a paper describing how it identified and stopped several users who were attempting to use its AI models for biological weapons research.
Anthropic noted that the cases outlined in the report are far from run-of-the-mill misuse, spanning everything from fake dating apps and surveillance tools to research tied to biological weapons.
Anthropic did not name the individuals or organizations behind the bioweapons-related research, nor did it specify which biological agents were involved. It did note, however, that some of the scientists involved had tried to disguise the true purpose of their work to slip past the company’s safeguards.
Another case in the report centers on a grant application for orthopoxvirus research submitted by a state-associated infectious disease laboratory, which outlined plans to handle live orthopoxviruses, a group that includes smallpox and monkeypox.
Anthropic said it decided to go public with its findings because it believes companies have an obligation to disclose misuse of their AI services, and warned that as models grow more capable, the risks they pose will keep climbing unless developers and defenders step up to address them.
The report’s publication follows a series of social media posts this week from Anthropic researchers who warned that AI could “kill all humans” by the end of the decade. Researcher Jacob Coxon, who resigned from Anthropic on Tuesday, said the warnings “were not a marketing stunt,” and that “no other human activity poses this level of danger.”
Anthropic explains that the more advanced research cases it stopped were only achievable because its newest models have become capable enough to carry out the work, something that, a year ago, would have required a team of skilled specialists.
Maybe you would like other interesting articles?