AI is No Longer Just an Assistant: Anthropic Report Unveils Alarming Cyber Security Threats

Kathmandu — While Artificial Intelligence (AI) continues to simplify our daily tasks, its misuse is rapidly turning into a major headache for security agencies worldwide. A comprehensive new report released by leading AI provider Anthropic has sent shockwaves through the tech industry.

Titled "Detecting and Countering Misuse of AI: September 2026," the 64-page report reveals that AI is no longer limited to being a simple assistant; rather, it is increasingly being weaponized. Covering incidents from December 2025 to August 2026, the report exposes how hacker groups linked to Russia and China are exploiting AI to execute massive cyberattacks, corporate espionage, and diplomatic breaches.

AI Agents: Autonomous Attack Networks

One of the most concerning trends highlighted in the report is the rise of AI Agents. Previously, hackers used AI merely for basic tasks like writing code snippets or drafting phishing emails. Today, through multi-agent systems, AI can autonomously divide complex tasks with minimal human intervention.

In this setup, one AI agent scans the internet for vulnerable systems, another develops targeted malware based on those vulnerabilities, and a fourth categorizes and exfiltrates the stolen data. Consequently, executing sophisticated cyberattacks on critical government infrastructure no longer requires heavy funding or large syndicates of expert hackers—a single individual with basic skills can now orchestrate entire attacks.

Targeting Diplomacy and Elections

The Anthropic report details how a Russian state-sponsored cyber espionage group, designated as GTG-20006, utilized Claude AI to spy on defense infrastructure in Ukraine and Europe. Notably, when security software blocked their malware, human intervention wasn't needed to fix the code; the AI agent autonomously analyzed the security blocks, refined the code, and re-launched the attack.

Similarly, a Chinese-speaking group active in Changsha, Hunan province—labeled GTG-10007—targeted over 50 international institutions across education, energy, health, and tech sectors. The report also highlights instances where AI was weaponized to influence elections in Malaysia using thousands of fake social media accounts, and to spread disinformation campaigns in Bangladesh and North Africa.

Allegations of Intellectual Property Theft Against Tech Giants

Beyond direct cyberattacks, the tech landscape is witnessing a surge in "Illicit Distillation"—the practice of querying a competitor's advanced AI model millions of times to reverse-engineer its reasoning capabilities and train local models.

Anthropic leveled severe accusations against Chinese tech giant Alibaba, claiming it carried out one of the largest AI thefts in history. According to the report, over a three-month span, thousands of fake accounts linked to Alibaba fired over 151 million queries into Claude to enhance Alibaba's proprietary 'Qwen' model. Several other Chinese firms, including DeepSeek, Moonshot AI, and Zhipu AI, were also cited for attempting to siphon capabilities from Claude's ecosystem.

Future Challenges and the Regulatory Debate

Anthropic clarifies that AI has not gone rogue on its own; rather, it has amplified human malicious intent, making cybercrimes drastically faster, cheaper, and more scalable. Future cybersecurity battles will fundamentally be fought as "AI versus AI." This escalating threat landscape has intensified global calls for stricter international regulations and oversight on advanced AI development.