The First Month of Dots: A Scorecard Against What OpenAI Promised

A launch keynote is a set of claims made under the most favorable conditions a company will ever get to make them. A month is enough time to check some of them against what actually happened. This isn't a verdict on whether dots is good — it's a plain accounting of what OpenAI said on September 29, 2026, and what's confirmed, partially confirmed, or still open as of October 22.

Held up: the sandboxing and permission architecture

OpenAI's core structural claims — a dot gets its own cloud computer and browser, separate from your machine unless you connect it; Custom Rules let you pre-approve, block, or require sign-off per category of action; Auto-Review checks consequential moves against your instructions and rules before they execute; proactive research stays read-only; an Activity View lets you watch and intervene — have not been contradicted by anything in the month since launch. No independent reporting checked across this series' research found evidence these mechanisms don't work as described. That's a genuinely good outcome for a month-one scorecard: the permission architecture, the part of the product OpenAI leaned on hardest to justify giving an agent standing access to your tools, held.

Partially confirmed: scale and reach

The claim that a dot can reach more than 4,000 applications through OpenAI's plugin ecosystem remains unchallenged but also not independently re-verified by anything in this month's reporting — it traces back to launch-week coverage, not a fresh confirmation. More pointedly, whether a dot can use a plugin outside that ecosystem is still explicitly unconfirmed per third-party testing. The "4,000 applications" figure describes OpenAI's walled garden accurately enough; it says nothing about whether that garden has a gate.

Unresolved: every pricing question OpenAI left open at launch

This is the largest category, and it hasn't moved in a month. Additional-dot and specialist-dot pricing remains unannounced. The usage grace period closes in late October with no published replacement structure. The closest thing to a hint — Finances in ChatGPT's expansion to Free and Go users and GPT-6.1 Sol's fraction-of-the-cost pricing sitting right next to the model that runs dots — is an inference, not an announcement. A month is a reasonable amount of time to still be finalizing pricing for a genuinely new product category. It is also long enough that "undisclosed" has stopped being a launch-week caveat and started being the actual current state of the product.

Unresolved: regional and enterprise rollout

The EEA, UK, and Switzerland exclusion for Pro-tier users is exactly where it was at launch, with no OpenAI statement explaining or timing a resolution. Enterprise, Edu, and Healthcare access remains admin-gated, with no public usage data on how many organizations have actually deployed it through the Agent 365 integration OpenAI positioned as the enterprise template at launch.

The pattern across a month of checking: the parts of dots that are hardest to fake — the sandboxing, the permission layer, the things a bad actor would need to break for a real incident to happen — held up under scrutiny. The parts that are easiest to leave vague — price, regional rollout, enterprise adoption numbers — are still vague, a month later, by what looks like choice rather than oversight.

What OpenAI never promised, and shouldn't be scored on

A fair scorecard also has to exclude claims OpenAI never actually made. OpenAI did not promise a specific price for additional dots at launch — it said more dots would exist, without committing to a timeline for pricing them, which makes "still unpriced a month later" a disappointment relative to user expectations rather than a broken promise relative to OpenAI's own stated commitments. The same goes for the EEA/UK/Switzerland exclusion: OpenAI never said when, or whether, that gap would close, so "still excluded" is consistent with what was said at launch, not a contradiction of it. Separating "OpenAI said this would happen and it hasn't" from "OpenAI left this open and it's still open" matters, because the first is a genuine accountability failure and the second is just an unfinished product — a less dramatic, more common, and more forgivable state for anything three weeks past launch.

The honest final grade

Call it a pass on architecture, an incomplete on commerce. OpenAI built the guardrails it said it would build and hasn't been caught contradicting its own safety claims in the first month — a genuinely meaningful result given how much of this series' skepticism, going back to the original side-by-side comparison with Grok Bot and Muse, centered on whether permission models would hold under real use. But one month later, compared against Grok Bot's Grokipedia takeover and Muse's frontier-model upgrade and growth-slowdown controversy, dots is also the product that has changed the least and revealed the least about its own business model. That's not a failure. It's a company that built the hard, trust-critical part first and is still working out how to charge for it — which, a year from now, may turn out to have been the right order to do things in.

What a month-two scorecard would need to show

If this exercise is repeated at the two-month mark, the items that would most change the grade are the ones currently sitting in the "unresolved" column: a published pricing structure for additional dots, a stated policy for what happens to usage counting now that the grace period has lapsed, a resolution one way or the other on third-party plugin access, and either an EEA/UK/Switzerland launch date or an explicit explanation for the continued exclusion. None of those require a new feature — they require OpenAI to finish answering questions it already raised by launching the product in the first place. That's a lower bar than shipping something new, which makes the continued silence on all four, a month in, the more notable data point than anything in the "held up" column above.

This closes the second batch of this series. What started as a three-week product collision in September has become a slower, more interesting story about which of these companies is willing to say the quiet parts — the price, the risk, the editorial responsibility — out loud. None of them fully has yet. That's worth checking again.

Part 47 of 47 in the ChatGPT Dots series. Previously: Dots vs Grok Bot vs Muse, One Month Later: What Actually Shipped. This is the final article in this batch. Browse more builder insights or explore AI skills at aiskill.market.