7 Ways Terrorists can Exploit AI
7 ways terrorists can exploit AI shows the real-world impact AI may have on terrorist’s ability. Changes include:
- Spoken passcodes for safety and security critical instructions to counter AI imposters
- Security vulnerability for risk assessments that use LLM
- Step change in terrorist propaganda
Click on the slideshow to view the SIRV visualisation →
Contents
1. Risk Assessments
Large Language Models (LLM) are increasingly used to assess threats in and around physical assets.
LLMs have a number of security flaws for example, prompt injections. A prompt injection can lead to LLMs ignoring alerts, releasing confidential asset locations etc.
2. Hackbots
Hackbots are AI systems that can autonomously hack:
- Identify vulnerabilities
- Select the best tools to use
- Identify previous successful attacks (a)
“If an enemy nation-state has a highly competent hackbot, that’s a serious national security issue.” Joseph Hacker
3. How to guides
AI can provide step-by-step guides on how to:
- Build weapons
- Optimise attack procedures
- Simulate attacks (b)
4. Misalignment
Developers of AI claim a lot of effort goes into ensuring AI models are aligned with humanity’s best interests. However, with open source models available, these models can be aligned to meet anyone’s aims, including terrorists. (a)
5. Critical systems
AI promises huge opportunities through interoperability of different systems. For example, Large Action Models execute workflows from a spoken instruction. Although critical systems will have high safety standards, linked non-critical systems may not.
6. Propaganda
Terrorists can use AI to:
- Translate text across languages
- Produce optical illusions that fool social media moderators
- Operate human like chat bots that give users the impression there’s support for extreme ideologies
- Rapidly generate huge amounts of new content (b)
7. AI is a great Imposter
AI enables high quality voice cloning. False instructions can be relayed by terrorists to misdirect and spread panic. (c)
References
(a) https://josephthacker.com/ai/2024/05/19/defining-real-ai-risks.html
(c) https://www.scientificamerican.com/article/a-safe-word-can-protect-against-ai-impostor-scams/
"SIRV helped us move beyond basic reporting into a system that actively supports decision-making". Les O'Gorman, Director of Facilities, UCB - Pharma and Life Sciences
