Google’s latest DeepMind research introduces an evaluation framework that systematically analyses how AI affects the effectiveness of existing attack techniques. The aim is to better prepare security teams for emerging threats.
Analysis of the impact of AI on attack techniques
The evaluation structure developed by Google DeepMind consists of four key steps.
- Selection of representative attack chains: Identification of typical attack scenarios, such as phishing, DDoS attacks or zero-day exploits.
- Bottleneck analysis: Identifying which steps in the attack chain can be improved by using AI, making them more effective.
- Developing benchmark tests: Creating benchmarks to measure AI performance in identified attack phases.
- Impact assessment: Estimating potential cost savings for attackers across the attack chain .
Tests conducted with the Gemini 2.0 model have shown that AI primarily increases the scale and speed of attacks. New attack techniques rarely appear, but existing methods become more effective through the use of AI. AI performs particularly well in tasks such as exploration, avoiding detection and maintaining access .
Implications for cyber defence
The conclusions of this analysis are crucial for organisations’ defence strategies:
- Increased effectiveness of AI in the installation and command and control communication phases: This requires the implementation of more advanced detection and response mechanisms for these stages of attack.
- Realistic penetration testing using AI: The framework can serve as a basis for simulating AI attacks to better prepare security teams .
Cyberinflation index as a new measure of risk
One of the interesting ideas put forward by Google DeepMind is the concept of a ‘cyberinflation index’. It would illustrate how AI lowers the economic barriers to carrying out cyber attacks, reducing the time and expertise needed. Such a measure could help quantify the risks associated with AI development in the context of cyber security .
The wider context of AI risks
Attention should also be paid to other potential risks arising from the development of AI.
- Use of AI to create biological weapons and bombs: There are concerns that advanced generative models could be used to design dangerous substances or devices.
- Manipulation of public opinion: AI can be used to create sophisticated disinformation campaigns, which poses a threat to democratic processes.
The development of artificial intelligence brings both huge opportunities and serious cyber security challenges. The assessment framework proposed by Google DeepMind provides valuable tools to understand and counter the impact of AI on attack techniques. However, continuous monitoring, adaptation of defence strategies and international cooperation on security regulations and standards are required to effectively protect against evolving threats.
