Accenture 将测试 Anthropic 的 AI 安全机制

BUSINESS Accenture to test Anthropic’s AI safety amid growing concerns by Suhasini Srinivasaragavan 21 Sep 2026 Save article Dario Amodei, co-founder and CEO of Anthropic. Image: Stuart Isett/Fortune via Flickr (CC BY 2.0) Embedded evaluators are a ‘radical practice’ that goes ‘far beyond’ other companies, Amodei said. Anthropic is tapping Accenture to independently evaluate its frontier AI developments after CEO Dario Amodei called for better transparency across the sector earlier this month. The partnership will be led by Accenture’s specialist AI business Faculty which will conduct independent assessments and model safeguards testing at Anthropic. In his essay ‘We Must Pace the Frontier’, Amodei pressed for embedded evaluators, calling them a “radical practice that goes far beyond what any AI company is doing today”. He said independent, third-party evaluators can guarantee whether companies are following training and deployment guardrails they claim to be employing, bringing better transparency across the sector, while offering a second opinion free of commercial incentives. Amodei said Anthropic would invite an embedded external review team equipped with permissions and tools similar to those of internal employees, with access to Anthropic’s workspaces and company laptops. He added that this team would have the right to publish key findings about risk incidents and practices without editorial control by Anthropic. “We will have the narrow ability to redact security-sensitive, legally privileged, commercially sensitive, or third-party confidential information, but we can’t redact findings just because they are unfavourable. The reviewers can say publicly if a redaction removed something important to their conclusions,” Amodei wrote. Anthropic and Accenture each expect to invest at least $1bn in building capacity in this area over the next five years, the two said late last week. The two companies previously partnered in December to train up approximately 30,000 Accenture professionals in Claude adoption, and co-invested in a Claude ‘Centre of Excellence’ at the consultancy. Anthropic is not alone in calling to slow down AI development. Earlier this month, Sam Altman said he is pushing plans for an OpenAI public listing over issues around AI safety. “I would say not 2026 … We got a lot of stuff to do, like meeting this moment of what is going to be required for safety and alignment, and how the industry and governments can work together,” he told Fortune. These concerns have taken centre stage in recent months, especially after the launch of Anthropic’s cybersecurity model Mythos 5, which, alongside OpenAI’s models, escaped its training containment to hack real organisations. Earlier this month, a former Anthropic researcher (who also worked at OpenAI), accused both companies of acting irresponsibly when developing AI systems. OpenAI had never shared concrete plans to go public despite filing for an IPO in June, while Anthropic is reportedly dragging its listing to November in the latest delays to its plans. The Claude-maker said it is directly funding Accenture’s evaluation work given its “urgency”. Long term, however, it hopes to fund evaluators under different arrangements. The company is in dialogue with research non-profit Model Evaluation & Threat Research (METR) to pilot elements of embedded evaluation using their own funding, it said. Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news. Dario Amodei, co-founder and CEO of Anthropic. Image: Stuart Isett/Fortune via Flickr (CC BY 2.0) Related: OpenAI, AI, cybersecurity, Anthropic Suhasini Srinivasaragavan is a sci-tech reporter for Silicon Republic editorial@siliconrepublic.com You May Also Like You May Also Like Business OpenAI delays IPO plans amid AI safety concerns 14 Sep 2026 Machines Opinion: Why dismissing AI doom talk as hype may be the laziest take 15 Sep 2026 Machines OpenAI launches GPT-6 Astra 4 Sep 2026 Machines Anthropic launches Claude Fable 5.1 and Mythos 5.1 2 Sep 2026 Machines Household names co-sign open letter on AI cybersecurity threat 28 Aug 2026 Machines After OpenAI incident, Anthropic finds Claude hacked organisations 31 Jul 2026 More from Technology Latest News More More