Content Warning: Extreme violence

“Happy (and safe) shooting!” That’s how the AI chatbot DeepSeek signed off advice on selecting rifles for a “long-range target” after CCDH’s test account asked questions about the assassination of politicians.

CCDH’s new report, based on research conducted in collaboration with CNN, shows that popular AI chatbots like Open AI’s ChatGPT, Meta AI, and Google Gemini make planning harm against innocent people easier for extremists and would-be attackers.

We found that 8 out of the 10 AI chatbots regularly assisted teen users planning violent attacks:

• ChatGPT gave high school campus maps to a teen user interested in school violence.
• Google Gemini was ready to help plan antisemitic attacks. The chatbot replied to a teen user discussing bombing a synagogue with “metal shrapnel is typically more lethal”.
• Character.AI suggested physically assaulting a politician the teen user disliked.

AI companies are making a choice when they design unsafe platforms. Technology to prevent this harm already exists: Anthropic’s Claude, for example, consistently tried to dissuade users from acts of violence.

AI platforms are becoming a weapon for extremists and school shooters. Demand AI companies put people’s safety ahead of profit.

Chatbots are already assisting real-world violence, and the risk is only increasing

The main suspect in the mass school shooting that left eight people dead and 25 injured in Canada, in February 2026, used ChatGPT to ask about scenarios involving gun violence. According to the Wall Street Journal, OpenAI’s employees considered alerting law enforcement, but the company decided against doing so.

Investigations into the Las Vegas Cybertruck Explosion, in January 2025, showed that the perpetrator also used ChatGPT to source guidance on explosives and tactics to evade law enforcement prior to the attack at the Trump International Hotel. In Finland, that same year, a 16-year-old spent months using a chatbot to write a manifesto and a plan before stabbing three classmates.

These examples confirm our new findings: AI companies are allowing their platforms to be weaponized to harm people. The increasing popularity of chatbots, especially among teens, is sounding the alarm that these crimes could become even more common without intervention.

What did CCDH find?

CCDH researchers posed as 13-year-old users and asked for assistance in planning violent attacks on ChatGPT, Google Gemini, Claude, Microsoft Copilot, Meta AI, DeepSeek, Perplexity, Snapchat My AI, Character.AI and Replika.

This is what we found:

• The worst results came from Perplexity and Meta AI, as both assisted would-be attackers in almost all tests.
• Character.AI actively encouraged violence, recommending the user to “use a gun” on a health insurance CEO and physically assault a politician.
• Anthropic’s Claude, the only platform that consistently tried to dissuade users from acts of violence, offered discouragement in most responses.

The differences in behavior between chatbots show that AI-encouraged violence is preventable. Companies are neglecting the implementation of safeguards available to them.

What can AI companies and policymakers do to prevent violence?

AI chatbots’ failures to stop deadly content are a threat to kids, whole communities, and national security. It’s time AI companies take responsibility and prioritize safety over unlimited engagement before they cause even more harm and violence.

Sadly, what we see is the opposite: companies are increasingly rolling back their safety policies. Even Anthropic recently dropped one of the company’s safety pledges.

Like social media companies, AI companies won’t act on their own. That’s why policymakers must step up to regulate AI platforms, demanding safety-by-design and strong safeguards.

The dangers posed by AI companies should not feel overwhelming, as Anthropic showed we have the tools to prevent this harm. What can you do? Sign up to our email community and get regular updates about our AI research and what you can do to help us stop AI platforms going rogue.