Intelligence is not authority.
Whatever the pace of AI progress, the conditions under which an agent’s actions take effect must be enforced outside its own reasoning.
A smarter model is not a broader mandate.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: https://t.co/OGyPb7yaYt
Of all yesterday’s announcements, this may be the one I still remember in two months :
Same $200. Roughly half the usage.
So, effectively ~2x the price.
Not exactly the kind of AI economics I was hoping for. :/
I’ll explain the new Pro 200 plan differently, before I start live tweeting from DevDay on things that are going out!
Today we are going to ship a number of things that increase what you can do across the Plus and Pro plans. A lot of compute is online for this increase. As we increase the floor, we are changing the relative difference between plans to be
Plus = 1X
Pro 100 = 5X
Pro 200 = 10X
and we are reopening subscriptions for Pro 200 (we had paused it). If you have an existing plan you will keep the 20X multiplier for a bit and also receive a lot of additional credits because we know changes are hard even if it means that everyone will get more in the end.
Independent evaluation is essential. A complementary principle: an agent’s mandate must be enforced outside its own reasoning, before actions take effect.
A smarter model is not a broader mandate.
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
Over the summer, we have been sprinting on safety priorities; it's more important than ever for capabilities and safeguards to advance together. We have more to do but have made a lot of progress. We are also going to be launching our next model soon.
There is an obvious tension here: on one hand, Astra is very good and we are excited to see what people will build with it. We are proud of our work.
On the other hand, we are clearly in a phase of development where we believe caution is warranted, and we are pacing our progress to ensure that we can meet the safety standards required by new capability levels.
Astra has been done training for a while now and is a significant step forward in both capabilities and alignment. For the models after that, we have been slowing things as needed to ensure that we can do sufficient work on safety and alignment.
AI is getting extremely capable; no one fully understands the consequences of this. Managing the transition to a world with abundant and powerful AI to optimize for safety and benefits to people should be one of the highest priorities in the world. It is our highest priority at OpenAI.
We have been living with the tension between being excited and anxious about progress for some time, and it is still discordant for us. We know it is much more discordant for other people. And yet, we believe strongly that the world needs to understand where AI is going and how models perform in the real world. More importantly, we believe the world will need aligned AI to manage the future phases of this transition.
An iterative loop where society and this technology evolve together is what will lead to the highest chance of getting this right.
So we hope you enjoy our new model, and we hope the world continues to take what’s happening in AI extremely seriously.
Tomorrow we will bring back the 5h limit for Plus accounts across ChatGPT Work and Codex. I had mentioned this a while ago, but then postponed it.
This is necessary as (a) the 5h limit allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage and (b) users on the Plus plan are relatively casual and new users, but then also just accidentally eat through their whole weeks usage and then are confused, making it not a great experience.
We are for the upcoming months keeping the 5h limit not enabled for Pro $100 and Pro $200 subscriptions.
Today, we’re introducing Echo: one adaptive model built entirely from a pool of open-weight models.
On our first internal task mix, Echo reached Fable-level results at roughly 1/3 of the total inference cost.
Try it: https://t.co/oIymERQqB2
Today, we are introducing Inkling.
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.
https://t.co/Ghebq5mG30
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
ACTION is the word everyone loves in San Francisco right now.
AI agents can search, write, call tools and update systems.
But production asks a harder question: who gave the agent the right to act?
Action is easy to demo. Accountable action is what gets deployed.
Show Codex a workflow once. Reuse it as a skill.
Record & Replay lets you show Codex a recurring task, like filing an expense report or submitting a time-off request.
Codex turns that demo into an inspectable, editable skill.
You control when recording starts and stops.
Boris Cherny of Anthropic on $NOW: "If I'm doing something and I don't have the context, I'm not going to do a great job... ServiceNow is a really a great way to bring in that context that it needs to do the job."
One of the best pieces of product engineering analysis I’ve read this year.
The key insight: the prompts aren’t documentation. They’re compiled product decisions. Every line reveals a tradeoff, initiative vs. overreach, memory vs. noise, speed vs. cache cost. Anthropic didn’t write instructions for an LLM. They wrote an operational philosophy.
The Claude vs. Codex framing is the most useful for builders. Claude pushes toward initiative. Codex pushes toward precision. Two different theories of what an agent should do when facing ambiguity, one says “move forward,” the other says “don’t drift.”
The deeper takeaway: prompt cache isn’t an infra optimization. It’s a design constraint that shapes observable product behavior. When token economics drive UX decisions, you’re doing cost architecture, not just engineering.
And for anyone building agentic systems: the value is no longer in the model. It’s in the decision layer that controls what the model does and what it doesn’t.
@JasonSCui@a16z published the clearest diagnosis of enterprise AI failure I've read from an investor.
Thesis: agents fail because they're blind. Build a context layer.
I agree. And that's why I wrote this.
Context tells an agent what to know. It doesn't tell it what it's authorized to do.
A bad answer is an error. A bad action is an incident.
Context layers are the prerequisite. Decision layers are the guarantee.
→ https://t.co/V80CZdF1mm