The AI that heals can also help kill
For years, the danger of artificial intelligence facilitating the creation of biological weapons belonged mainly to the territory of alarmist predictions. No longer. One of the most advanced companies in the sector, Anthropic, revealed having blocked five uses of its models related to high-risk biological research. It failed to demonstrate that those involved intended to cause harm, but identified attempts to access sensitive information about potentially deadly pathogens.
Anthropic is one of the leading companies in the development of advanced language models, competing directly with OpenAI and Google in the AI race. The revelation comes at a time when the technological race between large companies intensifies pressure to innovate rapidly, often at the expense of stricter security measures.
The case exposes a fundamental tension in the industry: the more powerful AI becomes, the greater the potential for misuse. The ability to generate synthetic biological code, model dangerous proteins, or suggest lethal chemical formulations represents an unprecedented risk. Experts warn that current protection mechanisms may be insufficient against malicious actors determined to circumvent them.
Regulatory authorities now face the challenge of keeping pace with technology that evolves faster than legislation can respond. Without clear legal frameworks and international coordination, the sector remains essentially dependent on self-regulation by companies, which leaves many questions open about the true magnitude of this problem.
