Anthropic 于7月1日重新部署Claude Fable 5，美国出口管制解除后新增网络安全分类器
英文摘要
Anthropic resumed global access to Claude Fable 5 on July 1, 2026, following the lift of US export controls on June 30 that had forced a suspension since June 12. The pause was triggered by an Amazon report showing a prompt that could bypass safeguards to identify software vulnerabilities and produce exploit code, though Anthropic demonstrated the technique was reproducible on many other models including GPT-5.5 and earlier Claude versions. A newly trained safety classifier now blocks that specific technique in over 99% of cases, rerouting blocked requests to Claude Opus 4.8 instead of refusing them outright. Anthropic also proposed a four-criterion jailbreak severity framework and plans immediate mitigations for the most severe class. During the pause, Zhipu AI released GLM-5.2, a 750B-parameter open-weight model with 40B active per token and a cost of $1.40/$4.40 per million tokens, offering a much cheaper alternative. Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, with access included for Pro, Max, Team, and select Enterprise plans up to 50% of weekly limits through July 7.
中文摘要
Anthropic 于2026年7月1日恢复了Claude Fable 5的全球访问，此前美国出口管制于6月30日解除，该模型自6月12日起曾被暂停。暂停源于亚马逊研究人员发现的一个提示词，可绕过安全防护识别软件漏洞并生成漏洞利用代码，但Anthropic证明该技术在GPT-5.5及早期Claude等多个模型上均可复现。新训练的安全分类器可在超过99%的情况下阻断该特定技术，并将受影响的请求路由至Claude Opus 4.8而非直接拒绝。Anthropic还提出了一套包含四个维度的越狱严重性框架，并计划对最严重级别立即部署缓解措施。暂停期间，智谱AI发布了GLM-5.2，一个总参数7500亿、单次激活400亿的开源权重模型，定价每百万token 1.40/4.40美元，提供了更廉价的替代选项。Fable 5定价为每百万输入token 10美元、输出token 50美元，Pro、Max、Team及部分企业版用户可在7月7日前享受每周上限50%的免费使用额度。
关键要点
Claude Fable 5 resumed global availability on July 1 after US export controls were lifted on June 30, ending a suspension that began June 12.
Claude Fable 5于7月1日恢复全球可用，美国出口管制于6月30日解除，结束了自6月12日开始的暂停。
The suspension was triggered by an Amazon report of a prompt that identified software vulnerabilities and generated exploit code; Anthropic found the vulnerability identification was reproducible on models like GPT-5.5 and earlier Claudes.
暂停源于亚马逊报告的一个提示词，可识别软件漏洞并生成漏洞利用代码；Anthropic发现该漏洞识别技术在GPT-5.5及早期Claude等模型上均可复现。
A new safety classifier blocks the reported technique in over 99% of cases, falling back to Claude Opus 4.8 instead of refusing the request, with increased false positives during routine coding.
新安全分类器可在超过99%的情况下阻断该报告技术，受阻请求回退至Claude Opus 4.8而非直接拒绝，日常编码时误报率有所增加。
Anthropic is drafting a four-criterion jailbreak severity framework (capability gain, breadth, ease of weaponization, discoverability) and will implement immediate mitigations for the highest severity class.
Anthropic正起草一套四维度越狱严重性框架（能力提升幅度、广度、武器化难度、可发现性），并将对最高严重级别立即实施缓解措施。
During the pause, Zhipu AI released GLM-5.2, an open-weight model with 750B total parameters and 40B active per token, priced at $1.40/$4.40 per million tokens, achieving competitive coding benchmarks.
暂停期间，智谱AI发布了GLM-5.2，总参数7500亿、单次激活400亿的开源权重模型，定价每百万token 1.40/4.40美元，在编程基准上取得有竞争力的成绩。
Fable 5 costs $10 per million input tokens and $50 per million output tokens, with up to 50% weekly usage included for Pro, Max, Team, and select Enterprise plans through July 7, then moving to usage credits.
Fable 5定价为每百万输入token 10美元、输出token 50美元，Pro、Max、Team及部分企业版用户可在7月7日前享受每周上限50%的免费额度，之后转为用量积分。