The models did not stop shipping because the market fell. The market fell because someone finally did the arithmetic on what shipping them costs.

That is the shape of this week, and Saturday is the right day to say it plainly, because nothing traded.

The evidence position

Evidence for this edition was reviewed through 23:15 UTC on Saturday, November 7, 2026.

Three limits belong at the top, and they are unusually load-bearing today. First, the day's supplied lead set for this edition was empty, so nothing here rests on a company announcement routed through the usual channel. Second, no filing, prospectus, registration statement or company press release was retrieved for the market, IPO or funding items below; those rest on secondary reporting, named in the evidence notes, and are labeled as reported throughout. Third, no intraday data is reported. Saturday has no session.

What follows separates what was closed and dated from what was told to reporters.

What the tape actually closed at

The last completed sessions of the week did the work. The S&P 500 closed Thursday, November 5, 2026 at 7,353.73, down 334.89 points or about 4.4% — a fall reported as erasing roughly $3 trillion in market value. The Nasdaq 100 lost about 6.4% that day, described in coverage as its worst session since the pandemic, and Nvidia fell about 8%.

Friday, November 6, added about 0.3% to roughly 7,376.86. That is a closing figure, not intraday, and it trims rather than reverses what reporting calls the worst week since April 2025.

Those are index levels with dates. They are not a diagnosis, and a single week is not a trend. Nothing in the record reviewed here establishes why buyers sold, though reporting points at data-center regulation risk after the midterm results and at unease about the debt funding the buildout. Both are attributions, not measurements.

The first thing repriced was a private company

The clearest evidence that this was a financing story rather than a capability story is that the biggest markdown landed on a company with no listed shares to sell.

Secondary reporting says prospective Anthropic investors are now discussing a valuation of $1.2 trillion to $1.3 trillion, with some weighing as low as $1.1 trillion — against the $1.4 trillion to $1.5 trillion previously communicated to bankers, and an earlier target of up to $2 trillion. One outlet reported figures as low as $1 trillion.

Nothing about the company's models changed across those two days. Something about the price of the money did.

On the other side of the same ledger, reporting says OpenAI has closed a first tranche of roughly $110 billion at a valuation of about $1.25 trillion, after earlier plans to raise as much as $300 billion at $1.5 trillion were scaled back by at least a third. Gulf investors including MGX were described as the largest participant, with SoftBank and Coatue in talks.

Treat every figure in those two paragraphs as reported. No offering document was reviewed.

The capability side kept moving anyway

None of that slowed the release calendar. OpenAI's GPT-6 Cyber shipped with the first Critical rating under its own Preparedness Framework and a reported 48.5% on ExploitGym against 32.1% for its predecessor, along with 81 previously unknown Firefox vulnerabilities found during testing. Access is gated to approved Daybreak Red customers.

And on November 6, Google DeepMind published AI-assisted cryptanalysis against legacy 1024-bit RSA: a signature-forgery technique reported as roughly 1,600 times more efficient against short-digest keys, and a batch-factoring method reported to cut the marginal cost of additional related keys by about 30 times. Coverage states the result was validated in a sandboxed simulation against an already-solved RSA-1024 benchmark key, that the padding weakness bites when an attacker controls roughly half or more of the signed block, and that keys of 2048 bits and above are not affected.

Read that last clause twice before forwarding the headline. The multiplier is a theoretical estimate, the primary technical report was not retrieved for this edition, and nothing here suggests deployed 2048-bit TLS is at risk. The interesting part is the target: not a benchmark, but an assumption the industry retired years ago and still runs somewhere.

Why a business buyer should care

The rest of this section is interpretation, labeled as such.

Cheaper capital was doing part of the selling. When money is free, a capability demonstration reads as a plan. When it is not, the same demonstration reads as an expense line. Buyers should expect more questions about cost per unit of work and fewer about benchmark position.

Valuation compression is a vendor-stability question. A supplier that has just repriced its future raises a legitimate due-diligence question about runway and roadmap, and a legitimate contractual question about what happens to your integration if the roadmap changes. That is a risk-register entry, not a prediction.

The capability announcements didn't get less consequential — they got less funded. A decade-old key length and a Critical-rated offensive tool both improve on schedule regardless of what the index does. The organization that defers security work because the budget cycle tightened has chosen the wrong line item.

Three questions follow. What is the evidence date on the market claim in your board deck, and is it a close or an intraday number? If a critical vendor repriced, what in your contract survives that? And which of your own security assumptions were retired years ago by a standard you still run somewhere?

Also reported, not verified here

Cohere closed a Series E at a reported CAD 35.7 billion valuation, about US$25.5 billion, with the Government of Canada committing up to CAD 1.6 billion. Agility Robotics began trading on Nasdaq as AGLT after a SPAC merger raising about $465 million gross, against $1.8 million of 2025 revenue and a $138.1 million loss.

Both are secondary accounts of company-reported figures. Neither was checked against a prospectus or filing for this edition, and the Agility revenue and loss numbers in particular should be read from the filing rather than from coverage before anyone cites them.

For context on what readers are opening

The most-read item in this publication was OpenAI Brought In an Outside Offensive Team. The Target Is the Seam Between Systems, at 19 counted visits. Those are visits, not unique people.

What remains unestablished

  • Any primary document — filing, prospectus, registration statement, press release — behind the Anthropic valuation range or the OpenAI tranche.
  • Whether the reported figures were discussed, agreed or signed.
  • Any causal link between the selloff and the IPO repricing. The sequence is documented; the causation is attributed.
  • The published DeepMind technical report, and therefore the assumptions behind the 1,600× multiplier.
  • Whether the reported Cyber scope-adherence figure holds outside OpenAI's own harness.
  • Any intraday price. Saturday had no session, and none was reviewed.

Evidence notes and limits

The index closes, the Anthropic valuation range, the OpenAI tranche, the Cohere round and the Agility listing figures come from secondary coverage and digests dated November 6–7, 2026, reconciled against one another where they overlap. They are attributed in the text and are not presented as verified against a filing. Where two outlets disagree — the low end of the Anthropic range — both are stated.

GPT-6 Cyber details come from The Verge's November 5, 2026 report and from later secondary summaries of OpenAI's own post. The model's Critical classification, ExploitGym scores and Firefox findings are company-reported; no external red-team confirmation was reviewed.

The DeepMind RSA result is described only as reported, including its own stated limits: a theoretical efficiency estimate, a sandboxed validation against a known-solved key, and no impact on 2048-bit and above. No technical report was retrieved.

The three questions, the framing of the week as a financing story rather than a capability story, and the vendor-stability point are this publication's interpretation. They are not a forecast, and nothing here establishes that any organization changed a decision or achieved an outcome.

The advertisement above is separate from this analysis and is not a source for any claim in it.