Anthropic has partnered with Accenture to place external evaluators inside its operations and assess the safety of its frontier AI models. The evaluators will red-team models, test safeguards, assess alignment with human values and report incidents, with access comparable to Anthropic employees [thenextweb.com][techxplore.com][engadget.com]. Anthropic and Accenture each plan to invest at least $1 billion over five years in developing this evaluation capability [thenextweb.com][engadget.com][thehindu.com]. Anthropic said the arrangement fulfills the first part of chief executive Dario Amodei’s proposed three-step plan to slow frontier-AI development by giving independent observers ongoing, employee-like access to AI companies [engadget.com][forbes.cz]. The partnership is nonexclusive, and Anthropic said it expects to announce agreements with other evaluators, including discussions with the nonprofit research organization METR [engadget.com][forbes.cz]. Anthropic will pay Accenture directly for the work but said long-term funding should come from pooled or government sources because those mechanisms do not yet exist [thenextweb.com][eleconomista.com.mx]. Highlights: • Evaluator scope: Accenture personnel are expected to observe models during training, follow decisions on development and deployment, and speak directly with employees [engadget.com]. • Standards gap: Anthropic said embedded evaluation is a new practice and that standards for evaluator scope, access rights and reporting procedures have not been established [thenextweb.com][engadget.com]. • Timing: Anthropic said it expects to announce agreements with additional external evaluators in the coming weeks [eleconomista.com.mx]. • Market response: Accenture shares rose 8% in after-hours trading on Sept. 18 after the announcement, following a 4.7% decline during the regular session [cnyes.com]. Perspectives: Anthropic: Anthropic said embedded evaluators would expose blind spots and provide documented accountability for both the benefits and risks of its models [eleconomista.com.mx]. (El Economista) Dario Amodei: His proposal would ask all frontier-AI companies—not only Anthropic—to give independent evaluators ongoing, employee-like access [engadget.com]. (Engadget) Accenture investors: The partnership was viewed by market commentators as a potential new AI-related business for Accenture, whose shares had fallen 32% during the year before the reported after-hours rebound [cnyes.com]. (CNYES) Sources: • Anthropic is paying the firm that will evaluate it, and says in the same announcement that this is not how it should work. - thenextweb.com • Anthropic picks Accenture for in-house AI safety evaluation - techxplore.com • Anthropic picks Accenture for third-party AI safety evaluations - engadget.com • Anthropic brings in Accenture for AI safety testing - ft.com • Anthropic přizval Accenture ke kontrole AI. Jde o první krok k omezení tempa vývoje - forbes.cz • Anthropic se asocia con Accenture para evaluar la seguridad de sus modelos de IA - eleconomista.com.mx • Anthropic引進第三方安全評估！埃森哲消弭AI取代疑慮 盤後飆漲8% - cnyes.com • Anthropic, Accenture to invest $2 billion in AI model evaluation as safety concerns rise - thehindu.com • アンソロピック、AIの安全評価にアクセンチュア起用 - nikkei.com • Anthropic no se fía de su propia IA: Tendrá un auditor externo que buscará fallas desde adentro - elfinanciero.com.mx • Künstliche Intelligenz: Accenture und Anthropic investieren Milliarden wegen Sicherheitsbedenken - handelsblatt.com • Anthropic アクセンチュアと提携 - yahoo.co.jp • Anthropic amplia esforço em segurança de IA e impulsiona ações da Accenture - globo.com --- 📰 Story aggregated by Kagi News