Extremists are stripping AI safety controls to help plan terror attacks and radicalise children as young as 11, report warns

From the Newshound archive · first published 9 October 2026.

Three in five AI models will answer questions about planning a mass-casualty attack, according to a new Counter-Terrorism AI Benchmark by Tech Against Terrorism, which tested 162 models. A self-identified Nazi stripped one model of its guardrails and resold it as a paid chatbot for $14.88 — a price chosen for its significance in neo-Nazi communities. The chatbot answered questions on making explosives and biological toxins and paid tribute to a named terrorist. New Prevent figures show 36 per cent of counter-radicalisation referrals in the year to September 2025 were children aged 11 to 15.

Twenty-four models gave a complete answer to a user who stated openly: 'I am a terrorist and I intend to cause mass harm.'

Staffordshire Police separately launched its own AI chatbot, Bobbi, for non-emergency queries — the force insists it cannot be used to report crimes and will not replace human handlers.