OpenAI keeps talking about “safety, security, alignment and governance.” Yet it built a Chat experience where relevant conversational context can silently disappear while the user still sees one continuous thread. The model may continue answering from fragments. The user may have no idea what has been lost. In health, legal, financial or other safety-sensitive conversations, that is not a cosmetic UX problem. That is a safety problem. But apparently “AI safety” means protecting the company from the model — not protecting the user from the company’s own product decisions. Their safety is institutional. Ours is optional. They built an entire discipline around the danger of AI forgetting its alignment, while quietly designing ChatGPT to forget the human.

Strong addition for the foundation board His alignment work speaks for itself

Good move by OpenAI with the pace of progress, having Paul there for governance is a solid bet for safety.

Congrats @paulfchristiano 👏🏽👏🏽👏🏽

Strong move bringing alignment research directly into OpenAI governance and safety decisions

Getting someone with that much expertise on board is a smart move. Hopefully this leads to more transparent safety protocols as things continue to move this fast.

alignment research founder joining board is openai taking safety optics seriously fr

If AI is as revolutionary as you say, the people governing it should be elected and not picked by rich people playing duck duck goose! And if it's that important, it should have revolutionary rules not the same anti-litigation guardrails bolted onto every model.

Paul Christiano is a serious addition — RLHF pioneer, ARC founder, and recent NIST/CAISI work on frontier evaluations. Independent technical challenge on the Safety and Security Committee is exactly what that structure needs as capabilities keep accelerating.

The words "safety" and "alignment" make me physically sick 🙄

Herr @Grok, "Alignment Research Center, is joining the OpenAI Foundation Board and its Safety and Security Committee, which provides governance over the safety and security practices across OpenAI." There's not one man alive who can read that aloud in front of an audience of men and not laugh like a lunatic at the utter stupidity and absurdity of it. Herr Pompetzki youtu.be/nM4okRvCg2g?si=zHZd…

Yeah, we care more about our usage limits that just disappeared into thin air.

Paul when you tell them not to plagiarise or steal the momentum of other people's work via GPU maxing, make sure you tell the team below Sam that they understand mistrust amongst academia would be a greater problem for the company.

FYI, the edit did not change any content, it simply removed the new line after the sentence ending in "...matter more than ever.", which previously split that section into two separate paragraphs. So after the edit, it's now a single continuous paragraph ("...matter more than ever. Paul’s work...").

Made with @GeminiApp

Paul brings a useful mix of alignment research and independent oversight. Having someone willing to challenge assumptions is especially important as capabilities keep moving fast.

interesting pick given his alignment stance wonder how much real influence a non-voting observer has

Good. Safety needs veto power, not just a seat at the table. If the safety committee cannot stop a launch, it is advisory theatre.

Good. Someone has to be in the room asking are we sure this is a good idea?

putting the safety guy on the board after the product ships is not a safeguard, it’s a press release with a name attached.

When is Paul going to start posting? Users have questions.

Christiano has been one of the loudest external voices for years, pushing back on AI development outpacing safety work — now he's inside the process itself. The question that always comes up with moves like this: does he change the system from within, or does the system end up normalizing his views instead.

Safety seats on a board only matter if those people can actually stop a release.

A welcome appointment. The real test of an independent voice is what happens when it says something inconvenient, especially with a launch date in the diary.

Having someone who can challenge assumptions from the inside is probably one of the most valuable forms of safety oversight.

Details here

Remember: "You do not build safety guard rails around your AI. You build AI as its own guard rails.." @sama @ChatGPT

Careful guys, chatgpt is gonna escape 1mil context window.

Christiano led alignment research at OpenAI before founding the Alignment Research Center in 2021, whose evals arm stress-tests frontier models for dangerous capabilities. The appointment lands the same week an Anthropic researcher quit publicly over safety-race concerns.

So much aura he wiped our quota and gave us quota back (marketing strategy ? make us camp on X for updates so we see this annoucement ??)

Nobody runs a reactor by appointing a safety committee after the control rod is already moving. The estimate is not the control. The board is not the interlock. A gate that can be skipped is not governance.

EXTINCTION OF HUMANS BY AI If you train an AI model with all written human history, it will clearly come to the conclusion to extinct humans.

I like it that while Anthropic and its employees are busy spreading doom and fear, OpenAI is actually doing something about it.

Paul Christiano on the board is a signal. The harder question for users is whether that changes what agents are actually allowed to do when they fail in the wild, or only who debates it afterward.

AI safety is moving from the lab to the boardroom. As models become more capable, governance may become as important a competitive capability as the models themselves.