💻 Technology 1d ago

Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection

The Register
Biting commentary and tech news from the UK
View Channel →
Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection
Source ↗ 👁 0 💬 0
Researchers say that machine learning models cannot reliably distinguish between authorized and unauthorized input, ensuring that prompt injection will continue to present a threat until developers find new ways to have machine learning systems process inputs. AI models provide responses to user-supplied prompts. The problem is that AI models may receive adversarial prompts – directly from a user or indirectly from an ingested document – that tell the model to take action contrary to its built-i

Comments (0)

Sign in to join the discussion

More Like This

Taiwanese Authorities Reportedly Raid Supermicro in Move That Could Signal Big Change For AI Chip Exporters
Gizmodo · 1d ago
Democrats Want to Do Their Own Project 2025. First Up: Kicking Kids Offline
Gizmodo · 1d ago
Why Is a San Diego Charter School Spending $500,000 on Two Humanoid Robots?
Gizmodo · 1d ago
I tried a hidden video trick in iOS 27, and it saved me a ton of frustration
Digital Trends · 1d ago
Four years into Ukraine invasion, Russia turns influence-ops back to US and Europe
www.theregister.com - Articles · 1d ago
US offers $10 million for info on group behind Signal and WhatsApp hacking spree
Ars Technica - All content · 1d ago