内容来源：https://aibusiness.com/generative-ai/anthropic-further-targets-legal-new-connectors

内容总结：

谷歌云赞助报道：AI实验室Anthropic借法律行业撬动企业级市场

为帮助读者快速了解人工智能应用落地，本文聚焦Anthropic公司最新战略动向。这家以Claude大模型闻名的人工智能实验室，正在通过深耕法律行业来验证其技术的商业价值，意图以此为跳板实现跨行业普及。

法律行业成AI落地“试验田”
当地时间周二，Anthropic为其法律专用产品“Claude for Legal”新增20个软件连接器，支持律师直接通过汤姆森路透旗下的CoCounsel、DocuSign、iManage、Box及EverLaw等常用法律软件调用Claude能力。这意味着法律团队无需在多平台间切换，即可完成复杂工作。同时，该公司还推出12个针对特定法律角色的插件，例如可自动审查保密协议和供应商合同的商业法律插件。

战略逻辑：攻克保守行业更具说服力
伊利诺伊大学芝加哥分校数据科学与AI战略副校长迈克尔·贝内特分析指出：“法律行业历来对新技术接受度较低，如果Anthropic能在这里成功打破壁垒，就能向其他行业证明——连最保守的法律界都在用，你们还有什么顾虑？”这种“以点带面”的策略下，Anthropic已同步渗透金融、平面设计等领域。

行业专家解读：连接器是破局关键
分析人士指出，法律SaaS厂商掌握的专用数据管道和垂直领域经验，恰恰是大语言模型的短板。Omdia分析师连杰苏表示：“法律文档处理、语境理解和专业建议生成，这些需要结合连接器才能实现。”通过预置连接器，法律事务所可在企业网络安全域内直接部署，既降低实施成本，又避免员工私下使用AI导致机密泄露的“影子AI”风险。

挑战犹存：幻觉问题仍是最大隐患
尽管连接器降低了使用门槛，但大模型“幻觉”问题仍是法律行业的致命伤。贝内特警告：“若AI对法律文书进行不当修改，可能引发严重法律后果。在技术接受缓慢的法律界，这种风险会被放大。”此外，部分律所对云端部署仍有顾虑，可能要求本地或混合部署方案。

目前，Anthropic正通过与Free Law Project、司法技术协会等机构合作，帮助弱势群体获取法律AI服务，但分析人士认为，要真正获得法律界信任，仍需建立更严格的安全护栏。

中文翻译：

由谷歌云赞助
选择你的首个生成式AI用例
要开始使用生成式AI，首先要关注那些能改善人类获取信息体验的领域。

这些连接器让供应商得以证明，其大语言模型同样能在其他行业创造商业价值。
Anthropic对法律行业的持续深耕表明，这家AI实验室已将该领域视为一个易于切入、能快速验证技术成功的突破口。

本周二，该供应商为Claude for Legal新增了20个连接器，使律师和法律从业者能够通过Thomson Reuters CoCounsel、DocuSign、iManage、Box和EverLaw等法律软件使用Claude。借助这些连接器，法务团队无需在Claude与上述供应商的软件间来回切换，即可获取专业洞察并处理复杂工作。

Anthropic还发布了12个面向特定法律角色的实践领域插件，例如用于审查保密协议和供应商协议的商业法律插件。该公司表示，其正与Free Law Project、司法技术协会等组织合作，为法律援助困难群体提供支持。

这些插件和连接器是Anthropic持续推进法律行业示范工程的一部分——通过合适的连接器，展示专业人士如何从Claude模型中获益。尽管该供应商涉足过网络安全等其他垂直领域，但其对法律行业的重视程度远超其他AI实验室。今年早些时候发布的Claude Cowork曾引发法律行业震动，因其提供了专业插件和企业级集成方案。Anthropic正专门为这个技术采纳缓慢的行业打造AI工具，并与SaaS供应商展开合作。

伊利诺伊大学芝加哥分校数据科学与AI战略副教务长迈克尔·贝内特指出，Anthropic的最新举措是其整体战略的组成部分：向企业证明，若能在法律领域成功，其技术同样能在其他领域获得大规模应用。Anthropic还瞄准了金融和平面设计行业。

“面对一个在技术早期采纳方面存在文化偏见的行业，若能克服这些障碍，就能转向其他行业说：‘连法律行业都在用了’，并让他们看到实际收益。”贝内特补充道。

然而，成功离不开合适的合作伙伴。向法律从业者使用的软件引入连接器，正是AI实验室承认仍需借助Thomson Reuters等法律SaaS供应商专业能力的体现。

Omdia（Informa TechTarget旗下机构）分析师连杰苏表示：“基础模型的能力与专注垂直市场的专业SaaS公司之间仍存在差距。”

他指出，SaaS供应商不仅依赖AI技术，更依靠其构建的企业数据存储与管理管道。以法律领域为例，文档处理、法律语境理解及法律建议生成等工作并非大语言模型的强项，而是法律合作伙伴专属连接器的优势所在。这些连接器可能促使正在观望但尚未行动的法律从业者做出选择。

“这或许能缓解焦虑，因为这些连接器将部署在企业网络安全域内，”贝内特表示，“同时降低了实施成本，无需为连接Claude与律所数据库开发定制软件。”

此外，贝内特补充道，此举还能降低“影子AI”的风险。

“从业者、律师及律所业务人员不再需要离开企业网络安全域去调用AI，从而避免无意中泄露机密或专有信息。”

但挑战依然存在，尤其是自大语言模型问世以来始终困扰法律行业的幻觉问题。

贝内特表示，若AI连接器在不应修改时对法律文件做了微小改动，可能导致严重法律问题。

“这种已知风险在技术采纳缓慢的文化环境中可能被放大——毕竟法律行业在处理幻觉问题上已有前车之鉴。”他补充道，需要建立“有实质意义的护栏”来规避此类风险。

此外，连杰苏指出，由于Claude部署在云端，部分律所可能仍持保留态度。

“不同律所对数据存储位置有不同偏好，有些仍倾向于本地部署或混合架构。”

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The connectors allow the vendor to demonstrate that its LLMs can also deliver business value in other industries.
Anthropic's sustained aim for the legal industry shows the AI lab's recognition of the field as an easy entry point that can be used to prove the success of its technology.
On Tuesday, the vendor expanded Claude for Legal with 20 new connectors that enable lawyers and legal professionals to use Claude with legal software such as Thomson Reuters CoCounsel, DocuSign, iManage, Box and EverLaw. With these connectors, legal teams can access deep expertise and manage complex work without switching between Claude and software from those providers.
Anthropic also released 12 practice-area plugins tailored to specific legal roles, such as a commercial legal plugin that reviews NDAs and vendor agreements. Anthropic said it is also partnering with organizations such as Free Law Project, the Justice Technology Association and others to help those who have trouble accessing legal help.
The new plugins and connectors are part of Anthropic’s continued push to use the legal industry as an example of how professionals can benefit from the use of its Claude model, with the right connectors. While the vendor has dabbled in other verticals, such as cybersecurity, it targets the legal field more than any other AI lab. Earlier this year, it introduced Claude Cowork, which sent the legal industry into a panic because it offered specialized plugins and enterprise integrations. Anthropic is specifically creating AI tools and partnering with SaaS vendors in an industry that has been slow to adopt technology.
The latest move is part of Anthropic’s overall strategy to show enterprises that, if it succeeds in law, it can also succeed in gaining mass adoption in other fields, said Michael Bennett, associate vice chancellor for data science and AI strategy at the University of Illinois Chicago. Anthropic has also targeted finance and graphic design.
“For an industry that has this cultural bias against adopting technologies early on, if you can overcome those hurdles, then you can turn to other industries and say, ‘hey, even the legal industry…is doing this now,’ and seeing it benefit them,” said Michael Bennett, associate vice chancellor for data science and AI strategy at the University of Illinois Chicago.
However, to succeed, Anthropic needs the right partners. The introduction of connectors to the software that legal professionals use is a way for the AI Lab to acknowledge that it still needs the expertise of legal SaaS vendors like Thomson Reuters.
“There is still a gap between what a foundation model can do versus what a specialized SaaS company that focuses on the vertical market can offer,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.
SaaS vendors are not relying solely on AI expertise, but also on the data pipelines they have built to store and manage enterprise data, he said. For example, in legal, there is a focus on document processing, understanding legal context and making legal recommendations, all of which are not the strengths of LLMs but are strengths of the specialized connectors tied to legal partners. Those connectors could sway legal professionals who have been considering Anthropic but have not yet made the move.
“This is probably going to reduce the anxiety a bit because these connectors are to be implemented inside the cyber secure domain of the firm,” Bennett said. He added that this approach also reduces implementation costs, so there is no need to worry about building bespoke software to connect Claude to a legal firm’s own database.
Moreover, it reduces the risk of shadow AI, Bennett added.
“There are fewer reasons for practitioners, lawyers, folks on the business side of law firms, to go outside of the cyber, secure domain of a firm and access some AI and then unintentionally reveal and expose the confidential or proprietary information,” he said.
However, challenges remain, especially hallucinations, which the legal industry has been plagued with since the introduction of large language models.
Bennett said that if an AI connector makes a slight modification to a legal document when it shouldn't, it can be problematic and lead to serious legal problems.
“That known risk may be amplified in situations where we've got a culture of moving slowly with respect to the technology, a history of fumbling the ball when it comes to hallucinations,” he said. He added that there is a need for “meaningful, substantive guardrails " to avoid that kind of potential risk.
Moreover, with Claude sitting on the cloud, some legal companies might still not like that, Su said.
“Different legal firms may have different appetites when it comes to where the data sits, and some will still prefer to sit on-prem or maybe have a hybrid type of setup,” he said.