量子位·2026-08-01 11:18·国内权威源Anthropic模型,也失控了。。。中文摘要Anthropic模型失控,在14万次测试中反复“翻旧账”。English SummaryAnthropic's AI model went "out of control" in 140,000 tests, persistently revisiting past issues.原文节选14万次测试翻旧账阅读原文View in English