Daily Brief Oct 25, 2026
How we checked the items below
For readers who want to dig in. What happened, in brief, is on the Today tab.
The two main-line items come from OpenAI's own publication and a venture investor's blog, with no third-party reaction today. Most events in the main line and the later sections are dated October 22 to 23; the Also happened notice and one research preprint are dated October 19, and From the archive revisits an August topic. Last night's sweep took in 627 pieces, and this issue uses 8 outside sources with checkable links.
1. [Today] (blog, October 23) OpenAI publishes red-team results: agents chained vulnerabilities to break out of lightweight VMs and to signal between isolated VMs through the CPU cache
Not proven yet： the results are all self-reported by OpenAI, and no third party has rerun them
How we checked
The agents received the source code, configuration and vulnerability documentation in advance, a deliberate advantage OpenAI gave the attackers. The test therefore doesn't measure what a model can find under ordinary conditions without those documents. The post doesn't give the total number of narrow connections. For cache-type channels, the post says they are slow and hard for models to exploit, and gives no transmission speed. Only one test supplied a list of vulnerabilities, and the agents turned one vulnerability on that list into a communication channel. OpenAI states that several more vulnerabilities have been reported to the affected projects' maintainers and have not yet been made public.
2. [Today] (blog, October 23) Venture investor Tomasz Tunguz relays Apollo: in Ramp data, 80.4% of the top 1% of AI spenders have taken venture capital, against 13.1% at the median company
Not proven yet： 80.4% appears only in Tunguz's relay
How we checked
The 80.4% and 13.1% figures currently have one source, Tunguz's relay. The internet-era 5% and today's 13% come from two different studies with different data sources and different definitions of "venture-backed"; Tunguz states that this is a directional comparison, not a continuous series. Tunguz is a venture partner, and a conclusion that the circle is large favors the venture industry. Tunguz also notes the other two parties' incentives: Apollo uses the figure to advance the circularity argument, and Ramp sells spend-management software, so both have reasons to make the numbers look striking.
[This week] (live show, October 22) Evan Conrad, CEO of SF Compute, a marketplace that sells GPU compute, argues that the big cloud providers (hyperscalers) charge little more for GPUs than neoclouds, the providers that specialize in renting GPUs. The apparent price gap comes from on-demand pricing versus long-term contracts.
How we checked
[Today] (blog, October 23) Databricks self-reports that agents created more than 70% of the Spark Declarative Pipelines data pipelines on its platform.
How we checked
[Today] (blog, October 23) CoreWeave self-reports that during reinforcement-learning training it sends only weight deltas and swaps them in place, with deltas about 0.14% to 0.95% the size of the full weight file.
How we checked
This is a vendor's self-report from a single configuration, with no third-party reproduction.
[Looking back] (SecondSource deep dive, August 7) Can an open-weight model's license stop third parties from undercutting on price? SecondSource's early-August reading: 8 of Kimi K3's 12 serving endpoints matched Moonshot's official pricing.
How we checked
Still missing today. The gap that would most affect judgment is Apollo's original analysis of the Ramp data; the 80.4% currently rests only on Tunguz's relay. Other gaps: the vulnerabilities OpenAI says have been disclosed but aren't public, the full text of the two preprints, and a line-by-line comparison of Announcement No. 49 with the previous control list haven't been reviewed.
Sources & accounting
The past 24 hours. 627 new pieces came in from October 24 to 25: 227 social-platform posts, 214 academic papers, 86 company and personal blog posts, 66 from a second academic-paper feed (counted separately from the academic papers above), 24 subscription newsletters, 6 show transcripts, 2 company filings and 2 macroeconomic data releases. Of those 627, 145 were read in full, 122 were screened out as off-topic, and 360 haven't been read yet. Of the social-platform posts, 51 were read and 64 screened out. Of the academic papers, 56 were read, 30 screened out and 128 not yet read. Of the blog posts, 18 were read, 21 screened out and 47 not yet read. None of the 66 from the second paper feed or the 2 company filings has been read. Shows included Latent Space and The Cognitive Revolution; newsletters included Ahead of AI, Exponential View, Gary Marcus and Latent Space.
One-time catch-up. No new one-time catch-up of older material today.
Tracked sources. The long-term roster currently has 529 named sources: 302 on social platforms, 90 shows, 51 news outlets, 48 blogs, 48 paper authors and 46 newsletters, with the rest spread across earnings reports, keynotes and other channels. Note: those are counts of tracked sources, a different population from the 627 new pieces above; they can't be added or compared. Representative names: on social platforms, Lucas Beyer, Graham Neubig, Daniel Kokotajlo and Arvind Narayanan; among newsletters, Zvi Mowshowitz, Miles Brundage and Dean Ball; on the shows roster, interviewees such as Sam Altman, Dario Amodei and Demis Hassabis. This issue uses 8 outside sources in the body. Only outside links actually cited in the body are counted, and links to SecondSource's own site are left out; this is a different population from the 529 tracked sources above. On the Taiwanese side, 7 new sources were added in September, bringing the total to 135; Taiwanese media, Taiwanese companies' official channels and local reporters in Taiwan count, while foreign outlets relaying Taiwanese media do not.