Terrorism, Extremism, Disinformation and Artificial Intelligence

The document examines how AI technologies—specifically generative AI, large language models (LLMs), and recommender systems—can be exploited to scale disinformation, impersonate individuals via deepfakes, and automate political micro-targeting. It contrasts these risks with AI's potential to detect harmful content and enhance counter-speech. The briefing also reviews emerging regulatory frameworks in the EU, UK, and US, and discusses ethical principles like accountability and 'Constitutional AI' as means to govern these technologies.

• Large Language Models (LLMs) can increase the scale and effectiveness of political influence operations by lowering costs and barriers to entry, enabling the creation of tailored messaging at scale that bypasses traditional detection methods based on copy-pasted text.
• Deepfakes pose significant risks to individual psychological safety and reputation, with a disproportionate impact on women, who are targeted more frequently than politicians by deepfake videos.
• The 'liar's dividend' describes a phenomenon where the existence of convincing deepfakes allows individuals to dispute the veracity of authentic content by claiming it is AI-generated, thereby eroding public trust and the rule of law.
• AI-powered political micro-targeting (PMT) leverages 'big data' and predictive models to shape voter behaviour through hyper-personalised messaging, raising ethical concerns regarding voter agency and the line between persuasion and manipulation.
• LLM-powered social bots are becoming increasingly difficult to detect as they move away from 'self-revealing' patterns and may evolve into autonomous agents capable of making independent decisions via APIs and search engines.
• Algorithmic recommender systems can create 'filter bubbles' that insulate users from opposing viewpoints, potentially damaging public debate and democratic institutions, although some research suggests the impact on political polarisation is contested.
• Highly capable foundation models present three primary public safety problems: the 'unexpected capabilities problem' (unpredictable dangerous functions), the 'deployment safety problem' (difficulty in steering behaviour), and the 'proliferation problem' (unregulated open-sourcing).
• AI provides critical opportunities for mitigating online harm, such as Meta's 'Few-Shot Learner' for detecting COVID-19 misinformation and the use of neural networks to detect deepfakes more accurately than human reviewers.
• The EU AI Act adopts a risk-based approach to regulation, but negotiations have faced deadlocks over foundation models, with France, Germany, and Italy advocating for self-regulation to avoid hindering domestic AI development.
• The UK's approach to AI regulation is characterised as 'pro-innovation' and 'light-touch', focusing on existing regulators implementing AI principles rather than a single holistic body of law.

APA

Gandhi, M. (2024). Terrorism, Extremism, Disinformation and Artificial Intelligence. Institute for Strategic Dialogue. https://www.isdglobal.org/wp-content/uploads/2024/01/Terrorism-extremism-disinformation-and-artificial-intelligence_A-primer-for-policy-practitioners.pdf

Chicago

Gandhi, Milan. Terrorism, Extremism, Disinformation and Artificial Intelligence. Institute for Strategic Dialogue, 2024. https://www.isdglobal.org/wp-content/uploads/2024/01/Terrorism-extremism-disinformation-and-artificial-intelligence_A-primer-for-policy-practitioners.pdf.

Wikipedia

{{cite report |last1=Gandhi |first1=Milan |title=Terrorism, Extremism, Disinformation and Artificial Intelligence |publisher=Institute for Strategic Dialogue |date=January 2024 |url=https://www.isdglobal.org/wp-content/uploads/2024/01/Terrorism-extremism-disinformation-and-artificial-intelligence_A-primer-for-policy-practitioners.pdf |access-date=23 October 2026 |via=Climate Insights Directory}}

BibTeX

@techreport{gandhi2024terrorism, author = {Gandhi, Milan}, title = {{Terrorism, Extremism, Disinformation and Artificial Intelligence}}, institution = {Institute for Strategic Dialogue}, year = {2024}, month = jan, url = {https://www.isdglobal.org/wp-content/uploads/2024/01/Terrorism-extremism-disinformation-and-artificial-intelligence_A-primer-for-policy-practitioners.pdf}, urldate = {2026-10-23}, note = {Indexed by Climate Insights Directory} }