OpenAI to let outside groups vet AI models earlier in development

OpenAI will let outside groups evaluate its artificial intelligence models for safety risks at earlier stages of development, the company announced in a blog post on Tuesday.

The ChatGPT maker intends outside organisations to conduct technical safety assessments during the training, evaluation and rollout of new models, Bloomberg Law reported, as part of an ongoing effort to address heightened concerns about the technology’s potential harms.

OpenAI laid out the priorities it sees for the reviews to work well, including “strong independence mechanisms, scientific rigor, robust security practices and clear responsibilities”, according to the report.

The move pushes scrutiny upstream: third-party evaluations would run across training and evaluation rather than only in the weeks before a launch, with several assessors expected to divide the work by expertise instead of a single team signing off on an entire model, ByteVyte reported. OpenAI has not yet named any partner and has not published access terms, the outlet said.

The announcement follows a pledge from Anthropic chief executive Dario Amodei and OpenAI chief executive Sam Altman to welcome third-party evaluators with employee-level access, and a letter from the AI Evaluator Forum — backed by AI pioneer Geoffrey Hinton — setting conditions including scientific objectivity, transparency and immunity from retaliation, The News reported. OpenAI is separately in talks with Anthropic and Google on a shared industry safety standards body, The Information reported.

Sources

• Bloomberg Law: “OpenAI to Let Outside Groups Evaluate AI Models at Earlier Phase” — https://news.bloomberglaw.com/artificial-intelligence/openai-to-let-outside-groups-evaluate-ai-models-at-earlier-phase
• ByteVyte: “OpenAI Expands Third-Party Safety Evaluations Into Model Training” — https://bytevyte.com/openai-expands-third-party-safety-evaluations-into-model-training/
• The News (Pakistan): “OpenAI, Anthropic open their doors, watchdogs want full access” — https://www.thenews.com.pk/latest/1416750-openai-anthropic-open-their-doors-watchdogs-want-full-access
• TechDefused: “AI rivals in talks to police their own safety standards” (reporting The Information) — https://techdefused.com/a/2l8vA7c/ai-rivals-in-talks-to-police-their-own-safety-standards