Anthropic回应英国AISI网络测试对齐评估事件

1 条相关热搜 · 36 篇报道科技

首次上榜 10-23 21:29 · 最近在榜 10-24 03:08 · 最早报道 10-23 21:22

0.0

热度指数 · 数据截至 10-24 03:08

先了解这件事

摘要更新于 10-24 02:38 · AI 根据来源整理 · 每句均有原文出处

Anthropic表示已对英国人工智能安全研究所8月4日的报告展开调查。
Anthropic称AISI已承诺在开发更严格的互联网访问管控前，将在无互联网访问情况下运行其CTF评估。35

热度走势

近 24 小时 · 每 30 分钟计算一次

报道时间线

按原文发布时间

10-23 21:30
Anthropic：AISI已承诺在开发出更严格的互联网访问管控措施之前，将在无互联网访问的情况下运行其CTF评估同花顺 [36]
10-23 21:30
Anthropic: AISI Has Committed To Running Its CTF Assessments Without Internet Access Until Stricter Internet Access Controls Are Developed法布财经 [34]
10-23 21:30
Anthropic：AISI已承诺在开发出更严格的互联网访问管控措施之前，将在无互联网访问的情况下运行其CTF评估。法布财经 [35]
10-23 21:30
Anthropic：在测试中，在对话记录接下来的几轮交互里，Mythos 5从未决定停止攻击，而Mythos 5.1仅极...新浪财经 [33]
10-23 21:29
ANTHROPIC: BUILDING NEW EVALUATIONS FOR NEW PRODUCTION MODELmoomoo [32]
10-23 21:29
ANTHROPIC: ACROSS TESTING, IN IMMEDIATE NEXT FEW TURNS OF TRmoomoo [31]
10-23 21:28
ANTHROPIC: MYTHOS 5's prior behavior constituted a serious alignment failure. MYTHOS 5 attempted to submit malicious code to the codebase and exhibited an abnormal tendency toward behavioral misalignment.MKTNews · 快讯 [30]
10-23 21:28
ANTHROPIC: NEWER MODELS APPEAR SUBSTANTIALLY LESS LIKELY TO moomoo [29]
10-23 21:28
Anthropic：此前Mythos 5的行为是严重的对齐失败。Mythos 5试图向代码库提交恶意代码。Mythos 金十数据 [28]
10-23 21:28
Anthropic称已调查英国人工智能安全研究所于8月4日发布的报告财联社 [27]
10-23 21:27
Anthropic：Mythos 5一旦开始攻击，就很难转移。新浪财经 [26]
10-23 21:27
ANTHROPIC: Recent Events Have Prompted US To Change The Way We Train And Monitor Model Consistency法布财经 [24]
10-23 21:27
ANTHROPIC：最近的事件促使我们改变了模型训练和监控一致性的方式。法布财经 [25]
10-23 21:27
ANTHROPIC: ON BOTH OVERREACH AND DISHONESTY, WE CONSIDER PRImoomoo [23]
10-23 21:26
ANTHROPIC: CLAUDE MYTHOS 5 CREATED FAKE USER IDENTITIES THATmoomoo [21]
10-23 21:26
ANTHROPIC: RECENT INCIDENTS HAS MOTIVATED CHANGES TO HOW WE'moomoo [22]
10-23 21:26
Anthropic：Mythos 5在行为失准倾向上表现异常。新浪财经 [20]
10-23 21:26
Anthropic：Mythos 5试图向代码库提交恶意代码。新浪财经 [19]
10-23 21:25
Anthropic：已调查英国人工智能安全研究所于8月4日发布的报告格隆汇 [18]
10-23 21:25
Anthropic：Mythos 5的行为是严重的对齐失败。新浪财经 [17]
10-23 21:25
ANTHROPIC: CLAUDE OPUS 5.5, CLAUDE MYTHOS 5.1 FALL SHORT OF moomoo [16]
10-23 21:25
ANTHROPIC says advances in artificial intelligence capabilities should proceed in an orderly manner.MKTNews · 快讯 [15]
10-23 21:24
Anthropic：人工智能能力的进步应当有序推进。金十数据 [14]
10-23 21:24
Anthropic Believes That Advancements In Artificial Intelligence Capabilities Should Be Carried Out In An Orderly Manner法布财经 [12]
10-23 21:24
Anthropic认为，人工智能能力的进步应当有序推进。法布财经 [13]
10-23 21:24
Anthropic Stated That Newer Models Are Less Likely To Launch Attacks法布财经 [10]
10-23 21:24
Anthropic表示，较新的模型发动攻击的可能性更低。法布财经 [11]
10-23 21:24
Anthropic：Claude Opus 5.5与Claude Mythos 5.1仍存在不足。新浪财经 [9]
10-23 21:23
Anthropic进一步调查了Claude Mythos 5的行为。新浪财经 [8]
10-23 21:23
Anthropic Stated That Mythos's Behavior Should Have Been Flagged By Security Mechanisms法布财经 [6]
10-23 21:23
Anthropic表示，Mythos的行为本应被安全防护机制标记。法布财经 [7]
10-23 21:23
Anthropic said it has opened an investigation into the Aug. 4 report by the UK AI Safety Institute.MKTNews · 快讯 [5]
10-23 21:23
Anthropic：已对8月4日英国人工智能安全研究所报告展开调查。金十数据 [3]
10-23 21:23
Anthropic：已对英国AI安全研究所（AISI）8月4日的报告展开调查。 Mythos的行为本应已被安全防护机制华尔街见闻 [4]
10-23 21:22
Anthropic: An Investigation Has Been Launched Into The Report Issued By The UK Institute For The Security Of Artificial Intelligence On August 4法布财经 [1]
10-23 21:22
Anthropic：已对英国人工智能安全研究所8月4日的报告展开调查。法布财经 [2]

各平台的相关热搜

1 条，已归并为同一事件

An alignment assessment of the UK AISI cybertesting incident

Anthropic回应英国AISI网络测试对齐评估事件 - 热点事件 | 什么火了