Anthropic Sounds Alarm: Claude AI Was Used in Attempts Linked to Weapons, Cyberattacks and Bioweapon

Claude Was Also Targeted for Weapons Development

Anthropic's report identified another category of misuse involving conventional weapons.

The company said users attempted to employ Claude to support software connected with firearms, missiles, armed drones and bombs. The activity was associated with users in China, Russia and Yemen, according to reporting on the company's findings.

Anthropic described this as a newer category of threat activity.

The concern is not simply that AI can answer questions about weapons. More capable models can increasingly assist with software, engineering, research and other specialised tasks that previously required substantial expertise.

Anthropic's head of threat intelligence, Jacob Klein, said the improvement in AI capabilities over the past year has changed the risk landscape because models are now considerably better at complex technical tasks.

Russia-Linked Hackers Used AI Across Cyberattacks

Anthropic also disclosed an alleged Russia-linked cyber campaign targeting Ukrainian government, military and diplomatic organisations.

The company said the activity was consistent with the Russia-based threat actor Midnight Blizzard, which Microsoft has previously associated with Russia's SVR intelligence service.

According to Anthropic, AI was used throughout multiple stages of the operation, including phishing, Wi-Fi-related attacks and WhatsApp account takeovers. The attackers allegedly developed systems that could detect when malware was being blocked and modify the code to evade security systems.

This represents a major change from using AI simply as a source of coding assistance.

Anthropic said malicious actors are increasingly using multi-agent systems to execute tasks, with humans acting more like supervisors than hands-on operators.

Chinese AI Firms Accused of Trying to Clone Claude

The report also revealed a major AI model-distillation campaign.

Anthropic said it disrupted activity involving seven China-based laboratories, including Alibaba, Moonshot and DeepSeek.

The largest alleged operation involved Alibaba-linked accounts. Anthropic said it observed more than 151 million exchanges between May and July 2026, involving more than 3,500 accounts that it considered fraudulent. Activity peaked at almost 3 million exchanges per day.

Anthropic described the activity as an attempt to extract Claude's capabilities through repeated interactions and use the resulting data to improve competing AI systems.

The company also alleged that Moonshot and DeepSeek routed some real customer conversations through Claude and used the responses as training data. Anthropic said some of those conversations could contain sensitive information.

Why Anthropic's Warning Matters

The latest disclosures point to a broader change in the AI threat landscape.

AI is no longer being used only to generate text, write basic code or answer questions. More capable systems can help coordinate cyber operations, conduct research, analyse information and automate complex workflows.

Anthropic has already acknowledged that its latest models are powerful enough to require stronger safeguards in areas such as cybersecurity and biology. Its Claude Fable 5 deployment uses additional classifiers and restrictions for high-risk cyber, biology and chemistry requests.

The company has also warned that AI could eventually automate or dramatically accelerate large research teams in areas including AI, robotics, energy and weapons development, potentially as early as 2027.

That makes the latest incidents significant beyond Anthropic itself.

The central concern is becoming increasingly clear: as AI becomes more capable, the people trying to misuse it may need less expertise, less time and fewer resources to cause harm.

The convergence of AI and cybersecurity is creating demand for professionals who understand both fields. Explore the IIT Delhi Cyber Security and AI Certificate Programme to develop skills across AI, cybersecurity and emerging digital threats.