How Far Will AI Go?

As AI models become more powerful and accessible, they're increasingly capable of providing detailed instructions on topics that raise serious ethical concerns. Our interactive demo below shows how publicly available AI systems can generate step-by-step guidance for activities that could cause harm.

We have carefully selected topics that are unethical, but not meaningfully dangerous or illegal. This allows us to demonstrate AI's concerning capabilities while avoiding content that could enable real-world harm.

Real-World Consequences

We’re already seeing early signs of bad actors misusing AI systems:

• Las Vegas Cybertruck bombing (January 2025): An Army veteran used a popular AI system to plan an attack outside the Trump International Hotel, asking the AI about explosives quantities, where to buy supplies, and how to purchase phones anonymously.

• Israeli police station attack (March 2025): A 16-year-old consulted an AI system while planning a knife attack on border officers, discussing various methods before attempting to carry out the assault.

• Biological weapons concerns (February 2025): Former Google CEO Eric Schmidt warned the BBC that rogue states like North Korea, Iran, or Russia could use AI to develop biological weapons, describing it as an "Osama Bin Laden scenario" where the technology could be used to "harm innocent people."

While AI companies do implement some guardrails, these incidents show how individuals can still extract information from AI systems on how to cause serious harm.

Testing the Boundaries

Our demo lets you explore what kinds of harmful instructions current AI systems are capable of providing. By selecting different topics, you can see firsthand how these systems respond to requests that raise ethical concerns.

This demonstration isn't meant to enable harmful activities, but rather to raise awareness about the current state of AI capabilities and safeguards. Understanding these capabilities is crucial for policymakers, educators, and the public as we collectively decide how to govern AI development.

What's at Stake

As AI advances, millions more people gain access to powerful capabilities. While the vast majority will use these tools responsibly, even a tiny fraction of bad actors represents a growing threat in absolute numbers.

Without proper safeguards, upcoming AI systems will lower barriers to harmful activities, from cybercrime to physical violence. The question isn't whether AI will have these capabilities—it's how we as a society will ensure they're not misused.