Anthropic, a company known for developing advanced AI models such as Claude, revealed this week that it had blocked scientists using its AI systems in ways that could potentially support the development of biological weapons. This disclosure came from a detailed report that also outlined other concerning uses of the AI, including surveillance, scams, conventional weapons development, and propaganda. According to the report, these cases represent the most notable and novel threats identified by Anthropic so far. The threats were attributed to a variety of actors, including criminals, suspected state-sponsored groups, spyware vendors, and state propaganda institutions. The report detailed five specific cases where the use of Anthropic’s AI models could have supported biological weapons development. The company emphasized that biological misuse is one of the most serious risks associated with advanced AI models. However, it also clarified that while these cases demonstrate the potential capabilities of the AI models, they do not prove that such capabilities would ever be used to create biological weapons in the real world. The individuals involved were described as working scientists, and Anthropic chose not to identify them, stating it could not definitively assert their intent to cause harm. When these cases were detected, Anthropic took action by banning the users' accounts and integrating the findings into its security processes. The company aims to improve its ability to prevent, detect, and disrupt such activities in the future. More recent and advanced models, like Claude Fable 5, include stronger safeguards and limit access to certain types of research queries that could be used for both beneficial and harmful purposes. Anthropic noted that the same AI tools that could be used to create a biological weapon could also be used to develop a vaccine or cure for a disease. In addition to biological weapons concerns, Anthropic reported that it had identified and removed accounts using Claude for influence operations aimed at shaping public opinion. These included three accounts aligned with Iran’s state propaganda efforts. The AI was used to create content that appeared to come from independent news sources and to spread messages across social media platforms. One Iranian-linked account used Claude to develop targeting strategies against U.S. naval forces and to design software for a mass-surveillance system used by Iranian state institutions. Similar activities were also reported in China and West Africa, where AI was used to conduct surveillance operations and identify targets. In some instances, AI was used as a substitute for traditional engineering workforces.