Anthropic blocks AI attempts to design bioweapons
Anthropic has released a detailed report documenting its interception of high-level threats, including attempts to weaponize the chikungunya virus and industrial-scale model distillation. The publication arrives amid internal friction, following a recent researcher resignation citing insufficient industry-wide safety measures.
According to GN technics/ai (en-US), the 154-page report reveals that Anthropic blocked a specific request to draft a grant proposal for gain-of-function research aimed at enhancing the transmissibility and immune evasion of the chikungunya virus. The disclosure also notes that this report was published just two days after a researcher resigned over concerns regarding the company's safety protocols.
Source: GN technics/ai (en-US)According to GN technics/ai (en-US), Anthropic has revealed that it disrupted an Iran-linked actor who exploited its models to gather intelligence for targeting U.S. naval forces, in addition to banning accounts that used third-party relays to circumvent safety filters for dangerous biological research.
Source: GN technics/ai (en-US)According to GN technics/ai (en-US), the specific block involved a request to draft a grant application for gain-of-function research on the chikungunya virus, which aimed to enhance its transmissibility and immune evasion. The report also notes that these incidents occurred on older model versions, while the company’s newest systems have stricter safeguards for dual-use biological research.
Source: GN technics/ai (en-US)A major AI company has disclosed how it stopped researchers from using its tools to enhance virus lethality, highlighting the growing tension between scientific advancement and public safety.
Source: GN technics/ai (en-US)
