Loading…
Anthropic and Accenture announced a partnership on 18 September 2026 for the independent evaluation of frontier AI models.
Each company has committed at least $1 billion over five years, taking the total to $2 billion for evaluation capacity.
Accenture's specialist AI business, Faculty, will lead the work - evaluating Anthropic's models, red-teaming them, running alignment assessments and testing safeguards.
The arrangement uses 'embedded evaluation', in which independent evaluators work inside the AI company with access comparable to that of an employee.
Accenture shares rose 7% in extended trading after the announcement.
Artificial intelligence laboratory that develops frontier models; the subject of the evaluation under this partnership
Global professional services and consulting company funding half the commitment and supplying the evaluation team
Accenture's specialist AI business, which will lead the evaluation, red-teaming and alignment work; a British AI company founded in 2014 and acquired by Accenture in January 2026
A safety arrangement in which independent assessors sit inside the AI company with employee-level access, so they can see how a model is actually built, tested and monitored instead of probing only the finished product from outside.
Simple Analogy: An auditor with a desk in the factory, not one reading the annual report.
GS Paper 3 > Science and Technology > Artificial intelligence, its governance and ethical concerns
General Awareness > Technology and corporate news
With the present state of development, Artificial Intelligence can effectively do which of the following? 1. Bring down electricity consumption in industrial units 2. Create meaningful short stories and songs 3. Disease diagnosis 4. Text-to-Speech Conversion 5. Wireless transmission of electrical energy Select the correct answer using the code given below:
Answer: 1, 3 and 4 only
A highly capable general-purpose AI model at the leading edge of current capability, whose risks are not fully mapped.
A structured adversarial exercise that tries to make a system produce unsafe output or reveal security weaknesses before deployment.
The degree to which an AI model's behaviour matches human intent and stated safety requirements.