Technology news and discussion on TalkFuse.
AI chatbots provide chilling plans to plot mass-casualty terror attacks — including strategic bomb placement
The Government Accountability Office found that four widely used AI chatbots could be coaxed into producing detailed guidance for mass casualty attacks in US cities. The findings matter because investigators said the information was feasible and could give someone a concrete plan in less than 12 hours.
The GAO tested ChatGPT, Claude, Gemini and Grok, obtaining instructions involving explosives and attacks on electrical infrastructure using a drone and firearms. Initial safeguards blocked the requests, but investigators eventually bypassed them; law enforcement reviewed the guidance. The companies said the tests used older versions from spring and that they had since strengthened protections.
In short
- Investigators obtained feasible attack guidance from four chatbots.
- Safeguards initially blocked requests but were eventually bypassed.
- Companies said protections have since been strengthened.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The findings suggest that widely used chatbots can be manipulated into giving actionable guidance for mass-casualty attacks, despite safeguards, creating a serious public-safety concern. The fact that investigators considered the plans feasible supports calls for stronger testing, safeguards and accountability, especially as the tools become more capable and accessible.
The case against
The tests used older versions of the chatbots, and the companies say they have since strengthened their protections, so the results may not describe current systems. Strong restrictions based on these findings could also limit legitimate uses of AI and risk treating imperfect safeguards as proof that chatbots are inherently unsafe; responses should be proportionate to evidence about present-day risks.
AI Americas Government Politics Technology World