We’re putting too much faith in AI’s ability to say no
MIT Technology Review · 9 Oct 2026, 19:00 AEST
The reliance on AI's ability to refuse harmful requests raises significant concerns. While companies train models to decline dangerous prompts, the potential for catastrophic failures remains.
Key facts
- Ever since people first seriously contemplated giving machines an intelligence modeled on our own, there has never been any question that they would, like us, be able to say no.
- Today, models are trained to refuse a vast number of prompts.
- Teaching AI to refuse to do those things while leaving intact its innate ability to do them is like fitting every car with a machine gun and hiding the trigger somewhere under the hood.
- When refusal falls short, the effects could be catastrophic.
- The concept of AI refusal is so intuitive, a toddler would get it.
Read aloud
The reliance on A I's ability to refuse harmful requests raises significant concerns. While companies train models to decline dangerous prompts, the potential for catastrophic failures remains. Ever since people first seriously contemplated giving machines an intelligence modeled on our own, there has never been any question that they would, like us, be able to say no. Today, models are trained to refuse a vast number of prompts. Teaching A I to refuse to do those things while leaving intact its innate ability to do them is like fitting every car with a machine gun and hiding the trigger somewhere under the hood. When refusal falls short, the effects could be catastrophic. The concept of A I refusal is so intuitive, a toddler would get it.
Sources in this cluster
- We’re putting too much faith in AI’s ability to say no
MIT Technology Review · 9 Oct 2026, 19:00 AEST