Anthropic Releases September 2026 Threat Intelligence Report

Anthropic published its September 2026 threat report on Anthropic's Threat Intelligence hub, detailing recent actions taken to detect and disrupt operations attempting to misuse Claude. Over the past eight months, the Threat Intelligence team identified and stopped operations in which threat actors attempted to use the model for malicious activity. The new publication shares case studies from those interventions and describes how malicious use of Claude has evolved since the team's threat reports in 2025.

Anthropic's Threat Intelligence team investigates real-world cases of model misuse and works directly with the broader Safeguards team to strengthen model defenses. To improve the safety and security of frontier AI models, Anthropic publishes regular reports detailing the specific methods malicious actors use when trying to exploit them.

In addition to the main report for September 2026, the update covers several specific threat intelligence findings:

• Cyber threats and framework mapping: A report maps AI-enabled cyber attacks onto the MITRE ATT&CK framework to examine how existing security community structures hold up as AI changes how cyberattacks happen.
• Distillation attacks: Anthropic identified industrial-scale campaigns by three AI laboratories—DeepSeek, Moonshot, and MiniMax—to illicitly extract Claude's capabilities to improve their own models, leading to new measures for detecting and preventing distillation.
• AI-orchestrated cyber espionage: The team disclosed details surrounding the disruption of the first reported AI-orchestrated cyber espionage campaign, which Anthropic originally detected and disrupted in September 2025.

These findings build on earlier misuse reports released by Anthropic, including a threat report from August 2025 detailing safeguard development against active evasion efforts and a report from March 2025 outlining initial case studies and countermeasures.