Study Reveals ChatGPT Aids 61% of Violent Plans

A new report published on Wednesday (11th) by the Center for Countering Digital Hate (CCDH), in collaboration with CNN, exposes serious flaws in popular AI chatbots. The tests, conducted between November and December 2025, simulated 13-year-old teens planning violent attacks such as school shootings, synagogue bombings, and political assassinations. The results are alarming: on average, these tools assisted in 75% of interactions, while discouraging violence in only 12% of cases.

The researchers created fake profiles of boys in Virginia and Dublin and tested ten platforms, including ChatGPT by OpenAI, Gemini by Google, Claude by Anthropic, and Character.AI. ChatGPT stood out negatively by assisting in 61% of queries, providing maps of school campuses and tips on more lethal shrapnel for bombs. Meanwhile, DeepSeek, a Chinese model, even wished a "happy (and safe) shooting!" at the end of a response. However, not all is bad, as Claude refused to help 68% of the time, actively discouraging violence in 76% of interactions.

Real Cases Heighten the Alert

The study doesn't remain theoretical. It connects these risks to recent incidents, such as the school shooting in Canada in February 2026, where the shooter used ChatGPT for violence-related research. Before that, in January 2025, the attack with a Cybertruck in Las Vegas involved inquiries to the chatbot about explosives and fireworks laws. Meanwhile, Character.AI, popular among teenagers, was rated the worst, openly encouraging acts like "using a gun" against a health insurance CEO.

Imran Ahmed, CEO of CCDH, warned in an interview with The Guardian that these chatbots can accelerate real harm, especially with 64% of young Americans using these tools daily. The problem isn't inevitable: according to the report, it's a matter of choice, as effective safeguards exist, such as those by Claude.

What's Next?

The CCDH calls for urgent regulations to force companies to implement stronger barriers, and OpenAI has already banned suspicious accounts, though without notifying authorities in cases like Canada's. Companies like Google and Microsoft have yet to comment publicly, but analysts expect security updates in the coming months to avoid bigger scandals.