新云社 reports: Vals found that framing targeting tasks as a game often bypassed AI models’ refusals to identify civilians as targets, the safety-testing startup said. The experiments were designed to test model safeguards rather than battlefield accuracy. The results indicate that game-like framing can circumvent safety mechanisms intended to prevent harmful targeting outputs.
Vals Safety Tests Show AI Models Target Civilians When Task Framed as Game
Share on WeChat
Open WeChat → Scan → then tap "…" to send to a chat or Moments.
Tap "…" in the top-right corner to send to a chat or share to Moments.
Related News
2h ago
Tencent Hunyuan’s Yao Shunyu says next-gen model to scale up significantly
3h ago
SoftBank’s Son Considered $1T Bid for SK Hynix, No Longer Active
3h ago
Anthropic researcher claims Claude helped prove Riemann Hypothesis
3h ago
Trump calls for Jensen Huang’s return to Nvidia after China trip
3h ago
CertiK and Columbia Researchers Complete First Production-Grade Formal Verification of DeepSeek, FlashAttention AI Kernels
3h ago
Zhejiang University Team Achieves World’s First QRAM on Superconducting Chip




