AI Models Are Revolutionizing Cybersecurity Threats

At the 2025 China Internet Conference, Zhou Hongyi, founder of Qihoo 360 Group, delivered a stark warning about the cybersecurity implications of large AI models. His analysis revealed how artificial intelligence is creating new vulnerabilities while simultaneously making cyber attacks more accessible and affordable for malicious actors.

The Hallucination Problem in Critical Systems

Zhou identified model hallucinations as one of the most significant security risks. "When faced with unknown information," he explained, "large models sometimes generate completely fabricated content." While this might be harmless in entertainment applications, Zhou emphasized the dangers when these systems are deployed in critical infrastructure, industrial production, or government operations.

Image source note: The image is AI-generated, and the image licensing service provider is Midjourney.

Democratization of Cyber Attacks

The cybersecurity expert highlighted how large models are lowering barriers to entry for cyber attacks. "Non-programmers can now use natural language to instruct AI systems to write malicious code," Zhou stated. He introduced the concept of 'injection attacks', where carefully crafted prompts can trick models into revealing sensitive corporate data.

Perhaps more concerning is Zhou's prediction that disgruntled employees could soon leverage these capabilities: "Even staff without technical skills might weaponize company AI systems against their employers."

The Rise of AI-Powered Cyber Warfare

Zhou painted a concerning picture of nation-state threats evolving with AI capabilities. "We're seeing hackers embedding their expertise into models, creating 'hacker agents'," he warned. With sufficient computing power, a single operator could potentially control hundreds of these AI agents simultaneously.

"The battlefield has fundamentally changed," Zhou declared. "Future conflicts won't just be human versus human - they'll involve algorithms fighting algorithms, machines battling machines."

Qihoo 360's Countermeasures

The tech executive revealed two key initiatives from Qihoo 360 to address these emerging threats:

1. Development of an AI security expert system for real-time threat detection and algorithmic defense
2. Creation of a specialized large model guardian to monitor model outputs and reduce hallucination risks

Key Points:

• Large AI models introduce new security vulnerabilities through hallucinations and lowered attack thresholds
• Natural language interfaces enable non-technical users to conduct sophisticated cyber attacks
• Nation-state actors are developing AI-powered 'hacker agents' that scale cyber warfare capabilities
• Qihoo 360 is responding with algorithmic defenses and model monitoring systems