Cisco Just Put This AI to the Test—and It Flopped
Cisco ran DeepSeek R1, an open-source AI model, through HarmBench, a brutal stress test with 50 harmful prompts. The result? It failed every single one. That’s like a security system unlocking the door every time someone knocks. This raises big concerns about how well open-source AI can handle safety risks. If even a big name like Cisco finds issues, what does that mean for the future of AI security?