Thursday, 17 September, 2026
Anthropic and OpenAI Plan Embedded AI Safety Evaluators, Raising Independence Questions
By TechShots Studio
Anthropic and OpenAI are proposing to give independent safety evaluators deeper access to their AI systems, including training processes and models. The evaluators could report safety incidents and assess whether systems are behaving as intended. Researchers broadly support the idea but say clear rules, legal protections, and genuine independence will be needed to prevent evaluators from simply operating under the companies’ control.
Read full story at TECHCRUNCH