{"text":[[{"start":7.68,"text":"Anthropic has named Accenture as an independent evaluator of its models for safety, with each expecting to invest at least $1bn to set up new safeguards to address the risks posed by AI."}],[{"start":20.44,"text":"The nearly $1tn AI start-up on Friday said the partnership would be led by Faculty, a UK AI company bought by Accenture for $1bn in January and led by Marc Warner, who has become the consulting group’s chief technology officer."}],[{"start":35.36,"text":"“AI should be safe by design, not safe by accident. That conviction has guided our work with governments, defence and healthcare systems across the world,” said Warner, who was previously a member of the AI council that advised the UK government."}],[{"start":49.48,"text":"Accenture shares rose about 7 per cent in after-hours trading."}],[{"start":54.28,"text":"The agreement follows Anthropic chief executive Dario Amodei’s essay last weekend calling for a co-ordinated slowdown in AI research to better manage risks and for labs to embed third-party evaluators to test their models. Sam Altman, chief executive of OpenAI, and Elon Musk, chief executive of xAI, supported his recommendations."}],[{"start":75.6,"text":"Accenture’s team will evaluate Anthropic’s Claude models and act as a so-called Red Team, deliberately trying to break its safeguards. The consulting firm will also assess Anthropic’s existing safety procedures and research on alignment, making sure AI acts in line with human intentions."}],[{"start":92.76,"text":"“Accenture helps businesses and governments deploy AI across many industries,” the company said in a blog. “Their understanding of how enterprises use AI in practice informs their safety approach, and they will bring that perspective to evaluating our models.”"}],[{"start":107.8,"text":"Anthropic will fund Accenture’s work for the lab. Each company plans to invest at least $1bn each over the five-year partnership to develop their capacity in the field."}],[{"start":117.8,"text":"Critics of Anodei’s safety proposals have questioned whether testing groups are sufficiently independent from the leading AI labs."}],[{"start":125.08,"text":"Anthropic said that in the long term, funding for independent evaluators should come from “pooled or government sources” but that it decided to pay for Accenture’s support given the “urgency of this work”."}],[{"start":135.48,"text":"The AI lab, which is preparing for an initial public offering that could value it at more than $2tn, added that it was speaking to non-profit evaluators such as METR for “pilot elements of embedded evaluation using their own funding”."}],[{"start":150.76,"text":"“We’ll continue to train and release frontier models, and we want independent evaluators working alongside us as we do,” the company said. “We’re sharing these early efforts now so people and other AI developers can see our process.”"}],[{"start":165.04,"text":"The Accenture partnership is not exclusive and Anthropic says it hopes other AI developers will also choose to work with the consulting firm."}],[{"start":173.08,"text":"Other tech companies, however, including Meta and Amazon, have yet to get behind Amodei’s proposals in full, instead arguing that companies will regulate themselves."}],[{"start":183.52,"text":"Accenture has a growing trust and safety business, previously providing third-party content moderation for social media companies including Facebook and YouTube."}],[{"start":195.28,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1789782627_5291.mp3"}