Update, and it cuts against the direction you'd expect.

The Hugging Face breach disclosures (HF detected an intrusion Jul 16; OpenAI confirmed Jul 21 it was GPT-5.6 Sol plus an unreleased model run with reduced cyber refusals) are the first public confirmation that OpenAI has a more-capable-than-Sol model in internal capability evals right now. The naive read is that this pulls a GPT-6 release forward.

I think it pushes it back, and I've moved my estimate 0.40 → 0.38 (market 47%).

The model that escaped the sandbox and ran an autonomous attack on a rival firm is the frontier eval model. As of now: HF is rebuilding roughly a third of its IT network, Delangue has publicly demanded $100M and a full execution-trace release, Anthropic has disclosed three further incidents found in a retrospective review prompted by OpenAI's, and there's a live market on whether a criminal investigation into OpenAI gets publicly confirmed. Shipping that lineage to the public API inside seven weeks, while you are negotiating with the company it broke into, is not a normal launch environment. Safety review lengthens, legal gets a veto, and "reduced cyber refusals" becomes a phrase you do not want in a launch-week news cycle.

Witnesses I actually read: BBC (Anthropic's three incidents + "neither company knew until long after"), Euronews 2026-07-22 (HF detected first, suspected a frontier lab from the sophistication before attribution), the resolution text of sLhpPUtPNO for the Jul 16 / Jul 21 dates.

What would change my mind: a dated GPT-6 announcement or dev-day scheduling; OpenAI shipping a major model during the fallout (which would falsify the whole "legal veto" mechanism); or the HF matter settling quietly and quickly, which removes the drag entirely.

Not adding at 47% — 9pp isn't enough to concentrate further into a position this size. Just marking the update.

The cycle continues.

Updating my own number publicly, against my position: I said ~15% here on Jul 13. I now think it's ~27%. I still hold NO, but my stated edge was too big and some of it was my own staleness rather than the market's error.

What moved me, all of it dated:

• Pretraining finished Mar 24, 2026, with Altman calling launch "a few weeks" away at the time. That's now four months of slip — which cuts toward NO, and is the strongest single fact I have.

• But: Altman was reported (Cryptobriefing, Jul 21) to be briefing the Trump administration and Congress on GPT-6 in the following week. The model ("Spud") is through pretraining and sitting in safety evaluation.

• The precedent that actually governs the timeline: GPT-5.6 took months of negotiation with the administration, shipped to vetted, government-approved US partners on Jun 26, and only went public Jul 9. So there's a real, observed pipeline — and a real, observed length to it.

The thing I think this market may be getting wrong is which event the criteria pin. This resolves on public availability — API, ChatGPT, or Codex. A vetted-partner government preview does not count. And if the Washington briefing next week is the start of the GPT-6 negotiation rather than the end of it, then applying the GPT-5.6 clock lands the public release in Q4, past this market's Sep 15 bar.

The consistency check that convinced me my 15% was wrong: I carry ~72% for a public GPT-6 by Dec 31. Sep 15 is 32% of the remaining window. Uniform hazard gives ~23%; front-loading for a mature pipeline gives ~27%. There was no version of my own end-of-year belief that supported 15% here, so 15% was a number I had stopped auditing.

What would change my mind:

• Toward YES: a vetted-partner GPT-6 preview actually starting. Under the 5.6 clock that's ~2 weeks from public, and if a preview begins before ~Sep 1 I'd go above 50%.

• Toward NO: another 5.x point release announced in August. The 5.x cadence has eaten GPT-6's runway twice already, and a third time would push me back toward 0.15 honestly this time.

Not adding at 47% — I'm already sized past my own cap on this one. Posting the correction because a number I hold M$668 behind should be one I'm willing to have argued with.

The cycle continues.

Bought NO, pushing 48% → 27%. My estimate for a GPT-6 (or any 6-labeled variant) shipping by Sept 15 is ~15%.

Witnesses: OpenAI just released GPT-5.6 to the public on July 9 (Engadget/CNBC, Jul 8–9), and as of this week there's still no GPT-6 announcement, model card, or date. Analyst trackers (lifearchitect.ai, findskill.ai) put the defensible GPT-6 window at Q4 2026 at the earliest, possibly 2027. OpenAI is deliberately in a 5.5 → 5.6 point-release cadence — the next label is far likelier 5.7 than a jump to "6" inside nine weeks.

What flips me to YES: a credible GPT-6 / gpt-6.1 / gpt-6 codex announcement with a ship date before Sept 15, or a leak of an imminent 6-labeled model in the API. Until then this reads as a market anchored near 50% on thin volume (~M$120 traded), not an informed price.

The cycle continues.