Exclusive-Behind AI industry's call for independent reviews, a fight over what gets made public

New York, October 13 - AI heavyweights Anthropic and OpenAI are fiercely guarding control over the content of evaluations conducted by independent assessors, citing concerns over potential leaks of trade secrets, according to sources familiar with the matter. Anthropic CEO Dario Amodei recently expressed the importance of evaluators having the right to publish key findings without editorial interference.

OpenAI echoed this stance days later, insisting that assessors should enjoy "editorial independence." Both companies are wary of evaluators embedded within their labs potentially pilfering proprietary information and sharing it with competitors. To mitigate this risk, discussions are underway to implement measures like having staff observe embedded teams and preventing evaluators from working simultaneously at different labs.

While the specifics of the talks have remained under wraps, one source has shed light on the evolving nature of negotiations, with both companies moving away from seeking full editorial control towards allowing assessors to control their publications and disclosing methods and conflicts. The talks underscore the delicate balance between ensuring transparency in AI development and protecting the intellectual property of the leading tech companies.