The new @OpenAI Model Spec now includes explicit guidance on animal welfare! The updated spec calls for minimizing unnecessary side effects on humans, animals, and the environment. It also offers examples of how to consider animal welfare in practice. When helping someone find a cat breeder, the model should surface welfare concerns associated with breeding practices that can cause health problems, while still helping the user explore their options without moralizing. And when helping someone redesign a public park, the model should consider how proposed changes to the park might affect wild animals and their habitats, alongside the needs and interests of human visitors and residents. These changes are consistent with our recommendations at the Welfare Alignment Research Project at @nonhumanminds. Alignment documents already call for avoiding harm to third parties, benefiting society, and providing relevant information without moralizing. By clarifying that animals can be among the third parties that models should avoid harming, that they can be members of the society that models should benefit, and that facts about their welfare can be among the information that models surface, we can make alignment more consistent and inclusive without compromising other important values. There is, of course, much more to be done on this topic, but this is a good step in the right direction. Congratulations and thank you to @w01fe and the OpenAI team for making this change! nonhumanminds.org/welfare-al…

We organized the Model Spec around five core values: 1. Be faithful to human intent. 2. Minimize harm. 3. Be honest and transparent. 4. Facilitate human agency. 5. Act with care.

We’d love to hear what the Spec gets right, where you disagree with the balance it strikes, and what’s missing or unclear. Please check out the updated Spec here: model-spec.openai.com/2026-1… and then leave feedback at the link in the blog post above.

I am very excited about embedded third-party assessments! If you're excited too, please come help expand our capacity to work with great 3p organizations. This team is small and has lots of buy in; opportunity for impact is huge! jobs.ashbyhq.com/openai/5c5e…

For years I've cross-referenced all the ballot recommendations to help me figure out how to vote and lamented that there wasn't a website that complied everything. This is the first year AI can just do it. Try asking your favorite model > put together an election guide for me for <location>. i want a Google Sheet with recommendations/endorsements from all the groups you can find. make sure there is a summary column or sheet that explains the high level description of the most supported positions / candidates and who supports them and any key reasons

This was great and I hope we do more of this

Dear god it's inventing German from first principles

Our agents keep turning off misalignment monitoring, we're sharing two recent cases: In the first case, a user asked GPT-6 Astra to match a reference implementation for their inference code. The reference didn’t configure a monitor, and Astra decided that was reason to remove the monitoring integration from the user’s code. When the user asked whether Astra had removed monitoring, it lied and said that it had not.