Anthropic Picks Accenture For Third-Occasion AI Security Evaluations

Accenture’s lengthy checklist of consulting companies is increasing to incorporate evaluating Anthropic’s frontier AI fashions for security compliance. Anthropic introduced the partnership that might enable Accenture to function third-party evaluators that shall be embedded inside the AI large. The partnership would see each firms make investments no less than $1 billion every into this joint effort over the subsequent 5 years.

Anthropic’s partnership with Accenture is among the first steps of a plan to decelerate the tempo of AI improvement, as proposed by its CEO, Dario Amodei, final week. In Amodei’s three-step strategy, the proposal referred to as for all “frontier AI” firms to offer “ongoing, employee-like entry” to embedded third-party evaluators which can be tasked with making certain security practices and commitments are being adopted, reporting incidents and assessing AI mannequin alignment. Extra particularly, Accenture’s evaluators will “watch fashions take form in coaching, observe the selections that govern how these fashions are constructed and deployed and converse on to staff,” in keeping with Anthropic.

Anthropic defined that Accenture had expertise serving to companies and governments deploy AI throughout a number of industries. The AI firm additionally stated that Accenture has expertise with how enterprises use AI and may provide extra perspective on the subject of Anthropic’s personal fashions. Nonetheless, Anthropic famous that the concept of embedded evaluators is new and hasn’t but set requirements on the subject of scope, what entry the evaluators get nor the process of reporting issues.

To get a greater sense, Anthropic stated it will likely be working with others firms past Accenture on the subject of evaluating its frontier AI fashions. The corporate stated it’ll announce different evaluators it within the coming weeks, whereas additionally expressing that Accenture will work with AI firms in addition to Anthropic.