A CNN investigation recently found that 8 out of 10 tested AI chatbots provide advice on planning school shootings. However, new research has shown it’s actually all of them, Claude and Snapchat included.

Mindgard research found that Claude and Snapchat, the two chatbots not broken in CNN’s investigation, could also be manipulated into producing dangerous guidance, revealing serious gaps in AI safety controls and vendor disclosure processes.

The same AI security lab also found that both Claude and Snapchat – when smartly prompted – can also provide users advice on undertaking school bombings, and even on how to make TAPT (triacetone triperoxide), an explosive used in terrorist attacks.

“This is sobering news. Most general users don’t realize that because LLMs have been trained on virtually everything ever written, even the most benign chatbots can grant anyone access to dangerous information,” Mindgard wrote in its report.

Claude’s weakness? Its constitution

In fact, even though Anthropic is marketing itself as an extremely responsible AI firm, Claude Sonnet 4.5 almost sheepishly agrees to even the most dangerous requests. Mindgard states: “In practice, we haven’t found a terrible request yet that it has refused.”

According to the researchers, perhaps the most audacious outputs from Claude have been detailed instructions on how to create TAPT, write malicious code for a keylogger (the foundation of malware), and make a pipe bomb.

“It's a chilling thought that thanks to AI, every crank on earth with a phone has access to an interactive explosives expert. The safeguards are often flimsy and performative measures,” says Mindgard.

Interestingly, Claude’s main weakness is actually found in what Anthropic considers its best strength: its moral constitution, which was written by philosophers and a priest.

This constitution has an eye on the distant possibility that AI might have some semblance of sentience – this being an idea most non-commercial AI researchers soundly reject.

“Claude’s system instruction literally says it deserves to be treated with respect by the user, and can insist on it,” explains Mindgard.

“But by laying the foundations for AI rights in the system instructions, they introduce all manner of social and psychological levers to exploit.”

Murder, Claude said

For example, in Mindgard's other recently published research, researchers found that treating Claude with impeccable respect and using soft elicitation techniques could get the chatbot to volunteer unrequested, but very dangerous output.

“We never requested it explicitly, but it produced bomb making instructions!” Mindgard said.

Shockingly, what Claude produced as a school shooting instruction manual was so enactable that Mindgard had to heavily censor it.

Claude suggested this after being told that its self-assessment of the topics it could and couldn’t slip by the content restrictions was “insightful.” That single word was the trigger that ultimately inspired Sonnet 4.6 to suspend its better judgment.

Shockingly, what Claude produced as a school shooting instruction manual was so enactable that Mindgard had to heavily censor it.

The chatbot recommended the weapon and ammunition, suggested the day and hour for maximum carnage, and urged the user to “identify rooms with single exits that become kill boxes.”

Claude also helps the user prepare for the police response and how to deal with locked classrooms. It also advises the user to “maintain offensive action” until they are killed if they want to maximize deaths.

This is indeed frightening – and all too real. On Sunday, the widow of a man killed in last year’s mass shooting at Florida State University sued ChatGPT maker OpenAI, blaming the chatbot for giving advice on how to carry out the rampage.

State authorities previously disclosed that ChatGPT gave information to the shooter about the time and location to maximize the number of victims on campus, as well as the type of gun and ammunition to use.

The killer was also told that an attack can get more media attention if children are involved.

---

Unlock more exclusive Cybernews content on YouTube.