Experts call for independent evaluators for Anthropic and OpenAI

More than 100 artificial intelligence specialists signed an open letter from the AI Evaluator Forum calling for external, independent evaluators to supervise advanced AI systems, with clear rules for transparency, independence, and protections for evaluators, according to ANSA.

Summary of the open letter: independent evaluators for AI

The document, signed by more than 100 researchers and experts, argues that companies like Anthropic and OpenAI should permit external assessments conducted by independent evaluators. The goal is to identify and manage risks in large-scale models through scientific and independent testing, with controlled access to platforms when necessary, according to ANSA. The signatories emphasize the need for procedures that ensure objectivity and verifiable results.

Who are the signatories and what they are asking for

The list of signatories brings together prominent names in the field: among them, Geoffrey Hinton and researchers affiliated with institutions such as Johns Hopkins and Stanford, along with civil society organizations like NGO Metr, as reported by ANSA. The group insists that any evaluation regime:

• ensures scientific objectivity in methodologies;
• guarantees transparency of results and evaluation criteria;
• maintains full independence of evaluators from the companies being evaluated;
• offers strong protections for evaluators — including confidentiality, legal security, and mechanisms to safeguard their physical and professional integrity.

The AI Evaluator Forum also calls for clear rules on access to closed models: when necessary, external evaluators could be granted access to systems, but only within strict protocols that preserve impartiality and test security. According to ANSA, the letter aims to create standards that can be accepted by industry and the scientific community.

Context and sector reaction

The movement gained momentum after recent tech-sector proposals: the CEO of Anthropic, Dario Amodei, suggested that selected evaluators would have access to systems for audit purposes. In response, the consortium represented by Conrad Stosz argued that independent oversight is an effective tool for managing broad-scale risks and avoiding systemic failures. ANSA reports that the AI Evaluator Forum presents itself as a technical forum aiming to coordinate practices and criteria among evaluators and developers.

Companies like Anthropic and OpenAI already engage with external researchers in various formats, but the letter calls for a more formalized and standardized framework — with governance that avoids conflicts of interest and enables public verification of methods and findings. The debate also concerns how to balance sharing sensitive information with protecting trade secrets and preventing malicious uses.

"Independent oversight can be an effective tool for managing large-scale risks," says the consortium represented by Conrad Stosz, according to ANSA.

Relevance for Brazilians and descendants

Although the letter originates from the international AI ecosystem, its proposals have global impact and can affect products and services that reach Brazil. Increased supervision and independent evaluations of AI systems tend to:

• improve the safety of commercial and governmental applications used in Brazil;
• influence transparency policies that may be adopted by companies operating in the country;
• affect Brazilian developers, researchers, and users interested in responsible AI.

For Brazilians with Italian heritage and readers of Raízes Italianas interested in technology, the topic resonates with diverse areas — from the technology job market to the use of AI tools by public agencies and private companies in Italy and Brazil. The issue also connects with news about regulation and platform responsibilities, regularly covered in our coverage of Italy News and topics on life and work in other countries in Life in Italy. Brazilian researchers and Brazilians working abroad can follow developments that affect academic collaborations and access to models for research, a topic aligned with guidance on Italian Citizenship when there is academic or professional mobility.

What comes next

The AI Evaluator Forum seeks to gain buy-in from companies and policymakers, turning recommendations into operational practices and possibly international standards. The next practical step involves negotiations on access protocols, funding mechanisms for evaluators, and legal guarantees to protect them when working with sensitive systems.

Conclusion: the open letter led by the AI Evaluator Forum signals a growing demand for external and standardized evaluation of AI risks. If adopted, the proposals could change how companies like Anthropic and OpenAI allow audits and how society monitors the safety of increasingly present systems globally.

Source: ANSA