The artificial intelligence sector is entering a critical phase of maturity, marked by a growing awareness of the risks inherent in its most advanced technologies. As the capabilities of models like Claude continue to evolve, Anthropic has chosen to break its silence by publishing a comprehensive overview of malicious usage attempts identified between December 2025 and August 2026. This report goes beyond warning of theoretical threats: it details concrete operations orchestrated by state actors, criminal organizations, and political influence groups seeking to divert AI from its intended purpose.
Among the key findings revealed by Anthropic’s "Threat Intelligence" team, the diversity of risk scenarios highlights the unsettling versatility of these tools. The documented abuses span seven critical domains, ranging from cyberespionage and malware creation to attempts at manipulating public opinion via propaganda campaigns. More alarmingly, the report sheds light on research aimed at designing remotely controlled weapons, as well as explorations in the biological field, including studies on viral transmission or toxin synthesis, illustrating the devastating potential of these technologies in the wrong hands.
The geopolitical implications are immediate. The company underscores that it has identified activities linked to states seeking to leverage AI to destabilize entire regions. Specific examples, such as the use of Claude to spread hostile content in complex international contexts, demonstrate how AI-generated disinformation is becoming a full-fledged component of hybrid warfare. These findings are forcing model developers to reassess their safeguards, as the line between technological innovation and global security risk becomes increasingly blurred.
Faced with these threats, Anthropic's message is unequivocal: urgent action has become an imperative necessity to secure the digital ecosystem. While the company notes that these incidents represent only a fraction of total misuse, they serve as a wake-up call for the entire tech industry. Moving forward, responsibility will no longer rest solely with developers; it will require close collaboration with regulatory authorities to curb the now nearly professional-grade exploitation of AI systems, where even API keys are becoming vectors for sophisticated attacks.