Anthropic published its September 2026 threat intelligence report on malicious use of Claude, detailing operations it identified and disrupted across cyberattacks, surveillance, fraud, influence campaigns and biological misuse over the past eight months.
The report says AI is increasingly moving beyond answering questions to orchestrating reconnaissance, phishing, malware development and data exfiltration with greater autonomy. Anthropic stresses that these cases are not representative of typical Claude usage and says the findings were used to strengthen its safeguards.