Experts warn that current AI safety mechanisms can fail and call for better safeguards against careless or unreliable AI outputs
Other Provisional 34% confidence first seen
Two technology outlets report commentary from AI and security leaders arguing that common fears about AI “turning on” humans lack scientific support. They instead emphasize that AI systems can produce unreliable outputs, including cases of hallucinations and the probabilistic nature of “refusal” behavior, and they call for stronger protections and safer usage practices.