Künstliche Intelligenz: KI-Firma: Aktivitäten rund um mögliche Biowaffen-Forschung

Anthropic, a leading artificial intelligence (AI) firm, has recently taken action against researchers attempting to utilize their software Claude for potential biowarfare research. This move comes as OpenAI, another competitor of Anthropic, highlighted the difficulty in determining the true intentions behind such requests, given the dual-purpose nature of such research in both medical and military fields.

The attempts, however, revealed that scientists from blocked regions with ties to state-run entities were covertly trying to access American AI developers' technology. While the exact research institutions and countries involved remain unnamed, Anthropic documented five cases of potential biowarfare research in a comprehensive report.

These researchers were not seeking direct scientific assistance but rather organizational support, such as drafting papers. Anthropic noted that AI models are generally programmed to deny the development of biological weapons, but in cases involving both medical and military applications, determining a clear boundary between harmless and harmful intentions can be challenging.

In one instance, researchers sought to use Claude to draft a proposal for Gain-of-Function research, which involves increasing the transmissibility of a virus - a method commonly used in lab environments to develop countermeasures, but also a potential route to weapon creation. Another case involved attempting to design a variant of bird flu transmissible among mammals using Claude's assistance, a move Anthropic emphasized could yield harmful knowledge useful for weapon development.

Lastly, a researcher was caught using Claude to create quarterly progress reports on toxin research.