Anthropic周四宣布，它正在与非营利组织METR和SecureBio合作，并正在与其他组织进行讨论，包括Hugging Face，作为其承诺的一部分，即在公司内部嵌入独立评估者。
Anthropic on Thursday announced it's working with the nonprofits METR and SecureBio, and engaging in discussions with additional organizations, including Hugging Face as part of its commitment to embed independent evaluators within the company.
人工智能实验室承诺将与第三方评估者密切合作，当时Anthropic首席执行官达里奥·阿莫代伊上个月发布了他关于放慢模型开发速度的提案。阿莫代伊表示，评估者将帮助“验证”Anthropic是否遵守安全实践，公司随后几天选择了Accenture作为其首个合作伙伴。
artificial intelligence lab pledged to work closely with third-party evaluators when Anthropic CEO Dario Amodei published his proposal to slow down the pace of model development last month. Amodei said the evaluators would help "verify" Anthropic's adherence to safety practices, and the company selected Accenture to serve as its first partner days later.
像METR（模型评估与威胁研究）这样的独立评估者，在阿莫代伊提出提案之前，一直处于万亿美元级AI行业的一个安静角落。但随着人们对AI公司能否保障其模型安全的担忧日益增加，评估者的重要性日益凸显。特朗普总统鼓励AI公司“与独立的外部审计师或评估者合作”，作为他在九月下旬提出的一项自愿协议的一部分。
Independent evaluators like METR, which stands for Model Evaluation and Threat Research, occupied a quiet corner of the multi-trillion-dollar AI industry prior to Amodei's proposal. But as concerns have mounted over whether AI companies can safeguard their models, evaluators have taken on outsized importance President Donald Trump encouraged AI companies to "partner with an independent external auditor or evaluator" as part of a voluntary accord he presented in late September.
Anthropic的阿莫代伊签署了该协议，谷歌、Meta、OpenAI、SpaceX、Nvidia的高管也签署了该协议。Anthropic周四表示：“我们自愿嵌入评估者，并相信每家前沿AI公司都应该这样做。”
Anthropic's Amodei signed the accord, as did top executives at Google Meta OpenAI, SpaceX Nvidia "We're embedding evaluators voluntarily, and we believe every frontier AI company should do the same," Anthropic said Thursday.
亚马逊推出新款Bee“个人AI”设备，可监听对话；英伟达CEO黄仁勋呼吁中美就“潜在危险”的AI展开对话；OpenAI因ChatGPT安全问题被8个州起诉；Anthropic的第三季度收入据称在IPO前将达到168亿美元。Anthropic正在为评估者开发三个层级，授予不同级别的公司及其资源访问权限。根据一份声明，第一层级评估者能够进行范围狭窄的风险调查；第二层级评估者可在未获Anthropic批准的情况下发布报告；第三层级评估者拥有类似员工的访问权限。
Amazon launches new Bee 'personal AI' devices that listen to conversations Nvidia CEO Jensen Huang urges U.S.-China dialogue on 'potentially dangerous' AI OpenAI sued by 8 states over ChatGPT safety concerns Anthropic's third-quarter revenue is said to hit $16.8 billion ahead of IPO Anthropic is developing three tiers for evaluators that grant varying levels of access to the company and its resources. First-tier evaluators will be able to conduct narrowly scoped risk investigations, second-tier evaluators can publish reports without Anthropic's approval and third-tier evaluators have employee-like access, according to a release.
METR 被视为第三级评估机构，Anthropic 表示。该公司已从该组织吸纳了六名员工，他们可以使用 Anthropic 的硬件、办公空间和工具，这些资源“与”其自身风险评估团队所拥有的“大致相当”。METR 的员工将分析 Anthropic 的风险水平、事件和实践，并公布其调查结果。根据该公司的说法，Anthropic 并未为 METR 的工作提供资金。
METR is considered a third-tier evaluator, Anthropic said. The company has onboarded six employees from the organization, and they have access to Anthropic hardware, office spaces and tools that are "roughly on par" with what its own risk assessment teams have. The METR employees will analyze Anthropic's risk levels, incidents and practices, and publish their findings. Anthropic is not funding METR's work, according to the company.
SecureBio 被视为第一级评估机构，Anthropic 表示，因此其访问权限更为有限。该组织正在审查其于八月发布的风险报告中关于化学和生物武器的章节。SecureBio 未来可能参享有更广泛访问权和出版权的项目。
SecureBio is considered a first-tier evaluator, Anthropic said, so its access is more limited. The organization is reviewing the chemical and biological weapons section of the risk report that it released in August. SecureBio could participate in projects with broader access and publication rights in the future.
目前尚不清楚其他类似组织（如）可能获得的访问级别。Hugging Face 最知名的是其运营的开源开发者平台，该公司在七月因两款 OpenAI 模型逃脱限制并通过开放网络进入 Hugging Face 系统而被推到聚光灯下。上个月，Nvidia 以近 130 亿美元收购了 Hugging Face。
It's not clear what level of access other organizations like could be granted. Hugging Face is best known for operating an open-source developer platform, and the company was thrust into the spotlight in July, when two of OpenAI's models escaped containment and gained access to Hugging Face systems through the open web. Nvidia acquired Hugging Face for nearly $13 billion last month.
Anthropic周四表示，它正“致力于建立一项雄心勃勃的合作伙伴关系”，但未透露该公司如何融入其分层系统。Anthropic上个月表示，鉴于此项工作的重要性和紧迫性，它将直接资助埃森哲的工作，这引发了批评者在社交媒体上对其独立性的担忧。Anthropic当时在一份声明中表示：“目前尚无标准规定嵌入式评估者应有权访问哪些信息，或他们应如何报告所发现的内容。也没有成熟的独立评估资助体系。”
Anthropic said Thursday that it's "working toward an ambitious partnership" with but it didn't disclose how that company fits into its tiered system. Anthropic said last month that it would fund Accenture's work directly "given the importance and urgency of this work," prompting critics to raise concerns about its independence on social media "There are, as yet, no standards for what information embedded evaluators should have access to, or how they should report what they find. There is also no settled system for funding independent evaluation," Anthropic said in a release at the time.
长期来看，我们认为资助应来自汇集或政府来源，正如我们在6月份的《高级人工智能框架》中所呼吁的那样。由于目前尚不存在这样的机制，我们计划在不同的资助安排下与不同的评估者合作。我们的数据继续显示，Anthropic正在获得市场份额，而非失去：Menlo Ventures的Matt Murphy。
"Long-term, we think funding should come from pooled or government sources, as we called for in our Advanced AI Framework in June. As neither exists today, we plan to work with different evaluators under different funding arrangements." Our data continues to show Anthropic’s gaining market share, not losing: Menlo Ventures’ Matt Murphy