Anthropic says it blocked multiple accounts that were using its AI models in ways that could support the development of biological weapons. The company’s new threat report outlines five real-world cases where users tried to bypass safeguards and hide the purpose of their research, including sensitive work involving infectious diseases, toxins, and possible gain-of-function research. Over one 30-day period, Anthropic identified about 35 separate research efforts with potentially concerning activity, though the company said it could not always determine whether the intent was harmful or legitimate scientific research. Anthropic also said it disrupted attempts to misuse its models for surveillance, scams, cyber operations, and conventional weapons development.
Why This Matters:
Anthropic’s report shows that people are already trying to push AI tools into dangerous areas, including biological research, cyber operations, surveillance, scams, and weapons development. These systems can help bad actors move faster, lower the skill needed to carry out dangerous work, and test the limits of built-in safeguards. When misuse is already being found and disrupted, it shows how quickly AI can become a public health, infrastructure, and national security risk. Biological threats, cyber risk, and system disruptions need to be treated as part of a serious preparedness plan.
Read the full article here.
Source: CNN
By: Deidre McPhillips
