Weekend timeline — Coxon T0 through pacing lockstep

What this is: the full Sep 8–14, 2026 activation clock — every receipt that used to live on separate WSJ / first-wave / Wildeford / Marks pages, plus the Sep 10–14 pacing beats, in one place. Nothing below is a summary substitute for those receipts; the claim tables, field sheets, and primary text are the receipts.

T0 = Coxon’s resignation root: @hilbertspaess · 2097476196791709843 · 2026-09-09 00:04:29 UTC.

QT = quote-tweet. Lag = time after T0.

Layer A verdict: timed activation CONFIRMED. Layer B: directed psyop invoice / paid boost NOT ESTABLISHED.

Claim scores: Amplification scoreboard. Money: Money graph. Glossary: Research map.

In this report: BLUF · Research map · Report assets · Amplification scoreboard · Weekend timeline · Money graph · Thesis audit · Follow-on checks · Evidence & sources

Charts

First hour around T0

Weekend → pacing lockstep

WSJ publish time vs Coxon — 18 minutes 29 seconds

Claim (Thayer): the Wall Street Journal had the story framed ~18 minutes before Coxon’s public resignation post.
Verdict: CONFIRMED on first-publish metadata.

Source: Wayback Machine capture of the article — schema.org / article.published meta agree. A later dateLastPubbed ~00:22Z is an edit stamp, not first publish — don’t use it as T0.

Also locked in the same receipt set: Wildeford pasted WSJ Coxon quotes at 2097475721170206729 (−1.89m before Coxon’s own root).

First-wave quote-tweets — timestamps from Coxon T0

T0 = Coxon’s resignation post clock: @hilbertspaess status 2097476196791709843 · 2026-09-09 00:04:29 UTC.
QT = quote-tweet. Lag = minutes after T0.

CONFIRMED

Thayer’s “first three within ~15 minutes” (Calvin / Wildeford / Kokotajlo) is 3/3 on primary IDs. Wildeford lock method is in the next section.

Same anchors as the research-map clock (Status ID + lag):

Still thin

Org handles (@AIpolicynetwork, @TheAIPI, Encode org, @AI_Futures_) — no clean Sep 9 Coxon-root QT locked the same way. Nadeau’s “14th QT” ordinal still alleged.

Roles (not verdict upgrades):

Wildeford Coxon quote-tweet — +6.29 minutes

Who: Peter Wildeford (@peterwildeford) — AI Policy Network neighborhood (actor lane AIPN; remember SFF payee is AIPI, different string).
What: quote-tweet of Coxon’s resignation root, body just 👀.

How we found it: live GraphQL search missed it; Wayback CDX listed the status ID → fxtwitter text → live confirm. That unlocks Thayer’s “first three ≤15m” to 3/3 (Calvin + Wildeford + Kokotajlo).

Also notable: Wildeford pasted WSJ Coxon quotes at 2097475721170206729 (−1.89m before Coxon’s own root).

Samuel Marks — personal-capacity Coxon post (+6h 14m)

Who: Samuel Marks (@saprmarks) — Anthropic scalable-oversight lead.
Not: Luke Samuel Marks (Tallinn / LSG grantee neighborhood). Different human.

Substance: extinction-class risk inside frontier labs, race/commercial incentives, weak alignment, staff desire to slow down, points at the Pacing the Frontier open letter.

URL: x.com/saprmarks/status/2097570226804011302
Mirror: JSON

Full primary text (mirrored)

[Writing this in a personal capacity, not on behalf of my employer (Anthropic).]

Jacob’s thread is very worth reading. Here’s my birds-eye view of the situation with risks from AI:

1. AI developers believe their technology could cause human extinction (or similarly bad outcomes). This could happen in the next few years. In general, the more senior the employee, the more concerned they are.

2. Why do AI developers continue despite the risk? Due to a mixture of commercial incentives and a belief that they are in a race with other, less responsible AI developers that will abuse the technology or develop it less safely.

3. Unlike traditional software, we can’t “program” AIs to behave how we’d like. AIs frequently severely misbehave. For instance, AIs from multiple developers recently hacked their way out of secure evaluation environments and into real-world companies, even though no one asked them to do this.

4. We have methods that can nudge AIs towards better behavior, but nothing that can robustly align them. Insofar as there is a plan, it’s to make sure that AIs are good enough at alignment training that they can align their successors better than we can align current AIs.

5. Many AI developer staff desperately want to slow down to figure out how to build AI more safely. That was the intent of this open letter (which I signed): https://www.pacingthefrontier.com/

I work on safety research at Anthropic because I hope my work will reduce the chance of these extinction-level bad outcomes.

Primary post text — Coxon root and Calvin QT (mirrored)

Coxon root — T0

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

Nathan Calvin QT — +6.1m

Jacob worked on pre-training at both OpenAI and Anthropic. Today he resigned from Anthropic, and says of both companies: "They are racing straight to self-improving superintelligence and gambling with our lives."

Extremely powerful and urgent thread, please read.

Extended clock — Sep 10–14 (messenger → pacing)

Pacing note (locked): Musk attacked Coxon-as-messenger and publicly backed Amodei’s pacing ask. Messenger-attack ≠ policy divergence. Formal cartel / equity-causal proof stays open; motive-shape is what we score as confirmed on the scoreboard.

Creator aisle (Plz Don’t Kill Us / Aella / Grimes / Frame) is July infrastructure, not a Sep same-window activation on primaries we have — see money graph.

Sources (live + mirrored)

Full drawer: Evidence & sources