In September 2026, AI developer Anthropic released a detailed report on instances where its AI model, "Claude," was misused. The report summarizes abuse cases across seven domains from December 2025 to August 2026: cyberattacks, influence operations, surveillance, fraud, biological misuse, weapons development, and adversarial distillation.
In the field of cyberattacks, the report points to "uplift," where AI increases the speed and scale of attacks. Specific examples include cases where groups suspected of Russian espionage used AI-driven workflows to automate processes from reconnaissance to data exfiltration. Additionally, trends such as "automated exploit generation," where AI is used for continuous vulnerability research and exploit development, were reported.
Regarding information operations, cases were confirmed where Claude was used to build fake social media profiles and news sites to conduct large-scale information manipulation. These included operations carried out in coordination with national elections and the mass distribution of AI-generated content to spread specific political views.
In terms of surveillance, instances were identified where AI was used to build malicious Firefox extensions and analyze vast amounts of data to identify targets. Anthropic stated that it has suspended the accounts identified through these activities and will utilize the gained insights to strengthen the model's safety measures.
Source: