CASP — Cambridge Programme on AI Science & Policy

Cambridge Programme on AI Science & Policy | Antonia Juelich

Artificial intelligence technologies, particularly generative AI and large language models, are increasingly leveraged by extremist groups to optimise propaganda, automate recruitment, and lower technical barriers for cyberattacks. This rapid technological democratisation enables non-state actors to scale operations globally with minimal resources, presenting unprecedented challenges for counter-terrorism agencies and digital platforms tasked with monitoring online threats.

To counter this evolving threat landscape, the Cambridge Programme on AI Science & Policy highlights the critical need for robust multi-stakeholder governance frameworks that bridge the gap between commercial AI developers and national security agencies. Current regulatory gaps allow malicious actors to exploit open-source models, bypassing safety guardrails to synthesise dangerous chemical or biological information and coordinate decentralised attacks. Consequently, international policymakers must establish standardised red-teaming protocols, mandate stricter API access controls, and foster global intelligence-sharing agreements to prevent the weaponisation of these emerging dual-use technologies.

Comment

The proliferation of open-weights models like Meta’s Llama-3 marks a fundamental shift in how non-state actors bypass traditional digital gatekeeping. Unlike closed-source APIs, these downloadable architectures allow extremist groups to fine-tune models locally for malicious code generation. This decentralisation neutralises the efficacy of centralized content moderation frameworks established under initiatives like the Christchurch Call. Consequently, the burden of prevention shifts from hosting platforms to hardware-level access controls and semiconductor export restrictions.

To read in details click here