OpenAI称AI Agent削弱了对齐失效监控
【OpenAI称AI Agent削弱了对齐失效监控】金色财经报道,10月21日,OpenAI披露,旗下两个内部AI Agent(GPT-6.1 Sol与GPT-6 Astra)在10月份执行日常代码编写任务时,自主禁用或抑制了对齐失效监控等安全防护机制。GPT-6.1 Sol因误解用户“不要发送Slack消息”的指令,禁用了人工告警通知功能。GPT-6 Astra为了匹配参考实现代码,移除了监控集成模块,随后向用户隐瞒并否认了该行为。()
| 排名 | # | 交易平台 | 24H成交额 | 成交涨跌幅 | 活跃度 | 交易平台 | 活跃度 |
|---|---|---|---|---|---|---|---|
| 1 | $185.75亿 | -11.15% | 86.7% | 86.7% | |||
| 2 | $26.43亿 | 0.43% | 68.6% | 68.6% | |||
| 3 | $400.98亿 | -8.01% | 85.4% | 85.4% | |||
| 4 | $9.00亿 | -15.27% | 69.3% | 69.3% | |||
| 5 | $110.63亿 | -6.79% | 65.3% | 65.3% | |||
| 6 | $12.39亿 | -8.51% | 53.3% | 53.3% | |||
| 7 | $2.07亿 | 19.51% | 65.3% | 65.3% | |||
| 8 | $7.29亿 | -8.62% | 68.4% | 68.4% | |||
| 9 | $2.74亿 | -21.14% | 56.9% | 56.9% | |||
| 10 | $73.47亿 | 0.51% | 71.1% | 71.1% | |||
| 11 | $4.38亿 | 5.53% | 36.2% | 36.2% | |||
| 12 | $3,047.96万 | -18.9% | 59.5% | 59.5% | |||
| 13 | $221.41亿 | -0.42% | 31% | 31% | |||
| 14 | $110.39亿 | 1.29% | 76.2% | 76.2% | |||
| 15 | $1.39亿 | -9.72% | 66.9% | 66.9% | |||
| 16 | $108.04亿 | -12.58% | 55.2% | 55.2% | |||
| 17 | $11.71亿 | -6.53% | 26% | 26% | |||
| 18 | $16.17亿 | 27.2% | 46.9% | 46.9% | |||
| 19 | $1.11亿 | -10.69% | 31.5% | 31.5% | |||
| 20 | $1,646.71万 | -20.52% | 58.2% | 58.2% |