English(EN) Anthropic disclosed it withheld cyber training from Claude Opus 5. The model still finds vulnerabilities nearly as often as the restricted Mythos 5, but complet

Anthropic因安全权衡而对Claude Opus 5隐瞒网络安全培训

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-25 14:15

Anthropic透露，他们故意从其Claude Opus 5模型中排除了网络安全培训。尽管如此，该模型在识别漏洞方面的能力几乎与其受限的对应模型Mythos 5一样频繁。然而，Claude Opus 5完成的漏洞利用数量明显减少，Anthropic正在公开记录这一权衡。 AI

影响这一披露凸显了Anthropic在平衡人工智能能力与安全性方面的方法，可能影响未来的模型开发和评估标准。

排序理由该条目详细介绍了与人工智能模型培训相关的具体研究发现和安全权衡。[lever_c_demoted from research: ic=1 ai=1.0]

在 Mastodon — mastodon.social 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

报道来源 [1]

Mastodon — mastodon.social TIER_1 English(EN) · schuler · 2026-07-25 14:15

Anthropic disclosed it withheld cyber training from Claude Opus 5. The model still finds vulnerabilities nearly as often as the restricted Mythos 5, but complet

Anthropic disclosed it withheld cyber training from Claude Opus 5. The model still finds vulnerabilities nearly as often as the restricted Mythos 5, but completes far fewer exploits. The company is publishing its safety tradeoffs explicitly. https://www. implicator.ai/anthropic-s…

链接 implicator.ai/anthropic-says-it-deliberat…

报道来源 [1]

Anthropic disclosed it withheld cyber training from Claude Opus 5. The model still finds vulnerabilities nearly as often as the restricted Mythos 5, but complet

相关实体

相关话题