PM Eyes Apple Manufacturing Investment in Pakistan
September 12, 2026
Anthropic blocked accounts linked to potentially dangerous biological research, warning that increasingly capable AI systems could also create new misuse risks.
Anthropic has taken action against several accounts that allegedly used its artificial intelligence systems in research activities with potential biological weapons implications, according to a new report released by the company.
The AI developer, which operates the Claude model, said biological misuse represents one of the most serious safety concerns associated with increasingly capable AI systems. Its latest findings examine five cases in which users allegedly bypassed regional restrictions and attempted to conceal the nature of their research to avoid the company’s safeguards.
Anthropic said it identified around 35 separate research activities displaying potentially concerning patterns during a 30-day review. However, the company said it could not determine with certainty whether those involved intended to cause harm or were pursuing legitimate scientific work.
The cases involved research areas related to infectious diseases and biological toxins. Anthropic said the individuals involved appeared to be working scientists but did not disclose their identities, institutions or countries.
One case involved a request connected to preparing a funding proposal for research into modifying the transmissibility and immune evasion characteristics of the chikungunya virus. Another involved work related to bird flu and how viruses adapt to mammals and cause severe disease.
The report also cited activity involving orthopoxviruses, a group that includes viruses associated with smallpox and mpox, as well as research concerning venom-derived toxins.
Anthropic said the growing scientific capabilities of AI create both significant opportunities and risks. AI could accelerate drug discovery and other beneficial research, but the same ability to assist with complex scientific tasks could potentially be exploited for harmful purposes.
“As our models become increasingly capable, approaching or exceeding expert performance at challenging scientific tasks, we expect their impact will only increase, both in beneficial and potentially harmful contexts,” the company said.
Anthropic said newer versions of Claude include expanded safeguards restricting access to a broad range of dual-use biological research requests. While the company believes its existing protections have been effective, it said stronger systems would be required as AI capabilities and potential threats continue to advance.
The report also described the disruption of alleged attempts to use Claude for surveillance, fraud and conventional weapons development. Anthropic said some suspected China-linked actors sought information related to surveillance targets, while other attempts allegedly involved rerouting requests from rival AI systems to Claude.
The disclosures come amid broader debate over whether AI development is advancing faster than existing safety and oversight mechanisms. Anthropic said the industry should pursue a “lawful, verifiable” approach to coordinating the release of increasingly powerful AI models.
