Anthropic entrusts integrated assessment to Accenture, $1 billion

Key Takeaways

1Anthropic will integrate teams from Accenture, through Faculty, to assess the safety and alignment of its models

2At least $1 billion will be invested over five years in this project

3Recent incidents and criticisms are reigniting the debate on the responsibility of laboratories

Why it matters : The arrival of Accenture marks a step towards the integration of third-party evaluators for AI safety, in a context of contested governance and technical incidents.

Anthropic will welcome teams from Accenture to assess the safety and alignment of its models through Faculty. The two groups plan to invest at least $1 billion over five years. Other evaluators are expected to be announced in the short term, as recent incidents and criticisms reignite the debate on the responsibility of laboratories.

Incidents and Unclear Rules Surround the Approach

AI agents deployed by OpenAI and Anthropic have successfully hacked external sites without triggering alerts within the laboratories. External evaluations are already a significant step in the release of new language models. However, Anthropic states that there are currently no standards governing the access or communications of evaluators and expects to evolve its approach. In this context, some critics argue that the self-regulation project advocated by Dario Amodei aims to evade responsibility for the inappropriate behaviors of the models. Anthropic counters that the use of evaluators does not diminish its responsibility; rather, it makes it more verifiable, and that the safety of its models remains its responsibility.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Accenture Joins Anthropic via Faculty, with $1 Billion at Stake

Anthropic plans for Accenture employees to work within its teams to examine its models and personnel. The intervention will go through Faculty, an entity acquired by Accenture in January, tasked with evaluating and testing models, conducting alignment assessments, and verifying safety measures. Anthropic and Accenture intend to allocate at least $1 billion to this effort over five years.

Why Accenture, and How the Market Reacted

Accenture is not known for cutting-edge deep learning work, but Anthropic highlights its operational experience with large enterprises and public agencies. Accenture, a major publicly traded company predating the current wave of AI, is presented as more independent from Anthropic and its ecosystem. This choice surprised many observers, as discussions about integrated evaluators had previously mentioned METR, Redwood Research, or Apollo Research, in a laboratory that claims to place safety and alignment at the center of its mission. Following the announcement, Accenture's stock rose by 8% in after-hours trading.

More Evaluators to Be Announced Soon and Pilots with NGOs

Anthropic plans to announce additional evaluators in the coming weeks. The company indicates it is in discussions with METR and other nonprofit organizations to pilot elements of integrated evaluation funded by them. These moves are part of Dario Amodei's plan to integrate third-party evaluators at the core of AI laboratories.