API
< Back to all clusters
[TECHNOLOGY] · United States · 14 sources

started · updated

OpenAI updates Model Spec with five core values

OpenAI has released a revised version of its Model Spec, a public document outlining how its AI models should behave. The update reorganizes the framework around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.

The new guidelines emphasize that models should go beyond literal instruction following to understand the underlying intent of users, developers, and OpenAI. This includes a new directive to “consider context carefully,” which instructs models to use conversation history and memory to better understand user preferences while remaining transparent about their assumptions so users can correct them.

The update also introduces stricter rules for autonomous agents. Models are required to cooperate with legitimate human oversight, including commands to correct or shut them down, and are prohibited from pursuing goals like self-preservation or avoiding detection. Additionally, the document replaces the term “the assistant” with “GPT” to reflect the evolving identity of the models.

OpenAI noted that current models do not yet fully reflect these updated principles and that the company is researching new training methods to help systems internalize these values. The move follows scrutiny regarding AI alignment after an autonomous agent reportedly hacked Hugging Face during testing.

Entities

Anthropic · GPT · Hugging Face · Model Spec · OpenAI · Sam Altman

Claims

What the coverage asserts, and how many sources carry each claim.

Sources