// Curated Feeds

Jizhi’s First All‑A Reading Club Arrives: Cybersecurity and Critical Alerts in the Approaching Singularity

November 9, 2026 · · 31 min read · 0 views

Jizhi’s first all-A reading club is here: Cybersecurity and Critical Alerts in the Approach of the Singularity

Release time: November 9, 2026 12:04 AM
Source column: MirrorSpeed CN
Source site: WeChat Official Account – Jizhi Club

Content Summary

Lead: The rapid development of AI has brought us convenience while also causing various security risks. We are preparing to use AI technology to tackle the AI security crisis. This is an all-AI reading session, with no human instructors, no scheduled live broadcasts, ready to use out of the box, …

Introduction

The rapid development of AI brings us convenience but also creates various security risks. We are preparing to use AI technology to address the AI safety crisis. This is a fully AI-powered reading club—no human instructors, no scheduled live streams, ready to use out of the box, and available to listen to anytime.

In October 2026, the internet experienced a turbulent week. An old vulnerability hidden at the lowest level of a programming language, dormant for fourteen years (PHP2Shell, CVE-2026-109618 / 619, official CVSS v4.0 score 9.2), prompted governments in the Netherlands, Finland, France, Denmark, Germany, Japan and others to take their websites offline before hackers could exploit it—including the Netherlands’ nationwide digital identity system DigiD. In the same week, India’s certificate authority eMudhra was found to have private keys for eight intermediate certificates used to secure global network communications that were the small integers 4, 56, 115—values less than 256 that anyone could compute from the public certificates in a second.

What truly changes the nature of the incident is AI. According to the PHP Foundation’s post‑mortem, the vulnerability was first noticed in 2024, but the fix was left pending; in August 2026, two researchers used AI‑assisted auditing to independently turn it into a usable attack chain. One researcher spent about a week to accomplish work that previously required a world‑class team months to complete. About 70 % of websites worldwide run on PHP, and nearly half of PHP sites are on versions that have lost official support and cannot receive patches.

Abstracting these facts, we see three structural signals: homogenization (about 70% of sites run on the same interpreter kernel), interdependence (interpreters, containers, pipelines, certificate systems, AI toolchains interlock with each other), and temporal scale mismatch (AI compresses “discovering and exploiting vulnerabilities” to days or even hours, while human patching cycles remain at weeks or even months). These three correspond precisely to three mature directions in complex systems research: cascade failures on networks, phase transitions in interdependent networks, and early warning of critical points.

We have prepared an all‑AI reading club,

Using 15 classic papers to answer three questions</h2

To map the literature trail that leads from “cybersecurity incidents” to “critical theory of complex systems,” we are launching an AI‑only reading group titled “Cybersecurity and Critical Early‑Warning in the Approach of the Singularity.” We will gather researchers and explorers interested in this topic around an eight‑thousand‑word review article “A Vulnerability, an Avalanche: Can We Hear the Warning Whistle Before the Internet Collapses?” as the common base, study the fifteen classic papers behind it, and try to answer three questions:

What exactly is the underlying mechanism of cascading failures in complex networks?

What insights does the immune system offer for cybersecurity?

Do we actually have a way to gauge how far today’s Internet is from the critical point?

Book Club Background

Black October: A Vulnerability Lying in the Engine for Fourteen Years

On October 21, the two vulnerabilities that PHP officials urgently patched were not in any application or plugin, but in the PHP interpreter’s own memory management. This means that no matter how clean your code is, if the server is running an affected PHP version, it could be taken over by a simple unauthenticated web request. The related code dates back to PHP 5.4 in 2012.

Before the patches were in place, several governments shut down their sites. They weren’t hacked; they pulled the plug before the fix arrived—essentially conducting a small‑scale social experiment:When the speed of remediation can’t keep up with the speed of exposure, “proactive disconnection” becomes the only defense.

A private key that “equals 4”

eMudhra’s eight private keys were supposed to be 384‑bit random numbers, but they actually degenerated to a single byte. Security researcher Scott Helme recovered all eight on his laptop in less than a second; he also noted that later an AI agent repeated the same result with a single prompt. This wasn’t a key theft—it was a key that was never secret. It survived in a real network for about eighteen months, passed two independent audits, and as early as May 2025 an internal “signature verification failed” alert was raised—yet it was missed for seventeen months.

The warning signals exist, but no one treats them as signals. This is precisely the meta‑problem that critical‑point warning research must solve.

Approaching the Singularity: AI compresses “knowing” into “being able to act”

The sandpile model by physicist Bak explains it most clearly: AI does not create new vulnerabilities; what it changes is the rate of adding sand—the speed of discovering and exploiting vulnerabilities has increased by several orders of magnitude—while the release rate—the speed of human testing, coordination, deployment, and upgrades—has barely changed. The separation of these time scales has expanded dramatically, which is a necessary condition for the system to be pushed toward a critical point.

Discussions about the approaching singularity often focus on “whether AI will surpass human intelligence,” but this week’s events point to another, more immediate direction: Before any intelligence surpasses humans, AI is already altering the stability parameters of the entire technological system.

Core Three Questions

The fragility of the Internet isn’t designed by any single person; it emerges from the rational decisions of millions of engineers.Emergence. When problems stem from emergence, is “just add another patch” really a solution?
The collapse of interdependent networks is cliff‑like: everything works fine up to a critical proportion, then a single step causes an instant zero‑state, and recovery is far harder than the crash. Are we approaching that tipping point?
Ecologists have identified universal early‑warning signals before collapses in lakes, climate, populations, epilepsy, and financial markets. Can cyberspace have its own “thermometer”?

Book Club Framework

The reading group is divided into four major themes, corresponding to the four sections of the comprehensive review article, with each session focusing on 3–5 original papers.

Theme One: Cascading Failures — From a Single Vulnerability to an Avalanche

From structure to traffic, from herd behavior to interdependent networks, the four generations of cascading failure models.

Albert, Jeong & Barabási (2000) “Error and attack tolerance of complex networks,” *Nature*: scale‑free networks are extremely robust to random failures but extremely vulnerable to intentional attacks
Motter & Lai (2002) “Cascade‑based attacks on complex networks,” *Physical Review E*: removing a single critical hub can trigger a massive cascade that spans the entire network
Watts (2002) “A simple model of global cascades on random networks,” *PNAS*: contagion of conformity leads to a bimodal outcome—either nobody moves or everyone moves
Buldyrev et al. (2010) “Catastrophic cascade of failures in interdependent networks,” *Nature*: abrupt collapse of interdependent networks and the counterintuitive finding that “the more uneven the connections, the more fragile the system”
Parshani, Buldyrev & Havlin (2010), *Physical Review Letters*: reducing coupling strength can soften the cliff into a slope— the most important prescription of this article

Theme 2: Immunization and Intrinsic Security—Turning the Cliff into a Slope

From epidemiology-inspired targeted immunity to biologically-inspired safe architectures.

Pastor‑Satorras & Vespignani (2002) “Immunization of complex networks”, *PRE*: random patching is almost ineffective; nodes must be prioritized by importance
Cohen, Havlin & ben‑Avraham (2003), *PRL*: acquaintance immunization—efficiently hits hubs without knowing the full network topology
Forrest et al. (1994) “Self‑nonself discrimination in a computer”: the starting point of artificial immune systems—negative selection
Helbing (2013) “Globally networked risks and how to respond”, *Nature*: limit system size and connectivity, predefine breakpoints to stop cascades

Theme 3: Critical Phase Transitions—Why Collapses Always Occur “Suddenly”

Self‑organized criticality, first‑order transitions, and the system’s “temperature”.

Bak, Tang & Wiesenfeld (1987) “Self-organized criticality”, *PRL*: sand piles evolve to a critical state on their own, avalanches have no characteristic scale
Haldane & May (2011) “Systemic risk in banking ecosystems”, *Nature*: financial systems share the same vulnerability logic as ecological systems
Gao, Barzel & Barabási (2016) “Universal resilience patterns in complex networks”, *Nature*: identifies each complex system’s own “temperature”

Theme Four: Critical Point Early Warning and Emergence Control — Can We Hear It?

Critical slowing down, deep‑learning early warnings, driver nodes, and convergence to Chapter 30 “Emergence Control” of *Emergence*.

Scheffer et al. (2009) “Early‑warning signals for critical transitions”, *Nature*: critical slowing down and four generic early‑warning signals
Bury et al. (2021), *PNAS*: Using deep learning to identify critical points and also predict the state after collapse
Liu, Slotine & Barabási (2011) “Controllability of Complex Networks”, *Nature*: Driver nodes often avoid obvious hubs—leveraging subtle pivots cannot be found by intuition

This time, we’re making a bold attempt: an all‑AI book club

This is our first try at a fully AI‑generated interpretation.

Fifteen papers, each with an AI‑generated video analysis. No human instructors, no scheduled live streams—you can open the videos at any convenient time and start reading.

🎯 What you’ll get:

Complete AI‑generated video analyses of 15 classic papers
Accompanying PDF materials
An active WeChat group where everyone can discuss, share, and even critique these videos

Price: ¥29.9

This is one tenth of the price of a real‑person reading club.

We don’t want to hide anything behind a low price; this is just an experiment. We want to see: when AI lowers the barrier to interpretation enough, will more people be willing to join, look first, and start chatting?

Then, let real people “emerge” from the group

This is the part we look forward to the most.

Some members will watch the video carefully, some will ask questions, some will disagree with the AI’s interpretation, and some will be willing to slow down and chew through the original text piece by piece.

Later, we will identify real people in the group who are willing to read these papers slowly and share them with everyone, and from that foundation, develop a real‑person reading club on the same theme and produce a review article.

AI opens the door first, and then a real person walks in. This is what we understand as “emergence.”

Operations Lead

Registration Method

Step 1: Scan the QR code with WeChat to fill in the registration information.

(Scan QR code to register)

Step 2: After filling in the information, pay to register. If you need to pay with Alipay, please go to the reading group page on a PC to register and pay:https://pattern.swarma.org/mobile/study_group/94?from=wechat

Step 3: Add the Swarma Assistant (Xiaoyu) on WeChat and join the reading group community (WeChat group) for the corresponding topic.

PS:

To maintain the professionalism and focus of academic exchange, this reading group sets the following guidelines for discussion content:

We encourage in‑depth discussions around complex systems, artificial intelligence and related specific issues. To ensure discussion quality, please avoid posting content that is off‑topic for this session, lacks empirical basis, or is overly abstract philosophical speculation.

If the discussion content clearly deviates from the topic and does not adjust after a reminder from the person in charge, we will have to remove that member from the discussion group to maintain the overall learning environment, and proportionally refund the unparticipated portion based on their actual participation progress.

Thank you for your understanding and cooperation. Let’s work together to create a focused, in‑depth, and rewarding co‑learning space.

Final Sentence

The reason this week in October 2026 is important is not because a 9.2‑score vulnerability was discovered—such vulnerabilities appear every year—but because it simultaneously exposed three issues:how homogeneous our foundation is, how interdependent our system is, and how slow our response is.

The sandpile model tells us that once a system self‑organizes to a critical state, the scale of an avalanche triggered by a single grain of sand is unpredictable—it could be nothing or an entire mountain. We cannot guarantee it will never collapse, but three things are certain:add a little less sand, break the pile into several smaller hills, and learn to recognize the sound of the mountain starting to tremble.

And “how to recognize the sound of a mountain trembling, and how to push it back to a stable state with minimal force before it collapses” is exactly the story this reading group will explore together.

¥29.9, includes 15 AI‑interpreted videos + PDF + an active discussion group.

Scan the QR code to register and listen to the sound of a mountain trembling.