ICHI Weekly Brief — 15 November 2026
Week in review: Cairn autonomous AI intrusions at scale, the wfx-loot GitHub pipeline campaign, ScreenConnect and N-central abuse, confirmed breaches at TSMC, Maersk and Baker Tilly, Europol's Operation Endgame, and the exploited flaws that matter.
Listen to this week’s podcast
MP3, 11.0 MB · · the full transcript is below.
The brief
ICHI Weekly Brief — 15 November 2026
The week's theme is automation at machine scale: an autonomous AI framework running real intrusions, a CI secret-stealer pushed to tens of thousands of repositories, and remote-management tooling abused across dozens of organizations in ninety minutes. Confirmed breaches, by contrast, lived mostly in third-party platforms and legacy systems.
Major cyber-health developments
- Cairn autonomous AI intrusions (Unit 42, Nov 12) — the week's lead and AI breakthrough: Chinese-speaking attackers used the open-source Cairn framework with Zhipu GLM 5.2 / DeepSeek V4 across 25 operations against at least 19 organizations (Aug–Sep 2026), with confirmed access/data at 10 — semiconductor supply chain, defense policy, civil society; >1M member records logged at one target. Moderate-confidence China-nexus assessment (cluster CL-STA-1238); likely one operator or a small team. Entry was via weak passwords and open endpoints.
- wfx-loot GitHub Actions campaign (Aikido, Nov 12): from Nov 10 ~12:45 UTC, stolen GitHub credentials across 4,837 compromised accounts pushed probe PRs ("docs: dependency audit notes") into 61,764 repos; the ~1,500-line secret stealer ran in at least 1,806 repos, exfiltrating Actions/cloud/AI-tool secrets, escalating via OIDC, and persisting through AWS trust-policy rewrites and Azure federated credentials. Packagist package digitalelvis/neuronai-studio 3.14.3 was also compromised.
- Japanese financial disclosures (Nov 12–13, reported Nov 14–15): AXA Direct Japan says unauthorized access to its MyAXA portal may have affected ~597,000 accounts (names, addresses, DOB, emails, phones; no passwords/card data in the system). Monex Securities says a third-party account-opening system may have exposed up to 1.13M email addresses and 1,592 identity-verification document images. Both are "may have leaked" maximums.
- Coles (Reuters, Nov 15): the Australian retailer is investigating a cyber incident and says it has found no evidence customer, team, or supplier data was compromised; operations continue.
Significant vulnerabilities and exploitation
- Barracuda ESG chain (confirmed exploited): pre-auth authentication bypass (CVE-2026-116606) chained with OS command injection (CVE-2026-111560) to root. Barracuda (Nov 9) confirmed exploitation affecting a small number of organizations; auto-pushed fix 9.4.0.030; CISA KEV deadline was Nov 13 — verify the patch landed.
- OpenSSH "SSHhole" CVE-2026-123158 (status change to confirmed exploitation): the musl/no-PAM long-password bypass moved from scanning (SANS honeypots: 23 attempts Nov 1 → 713 Nov 6) to confirmed post-exploitation — password-hash and root-key reads, authorized_keys modification, and a Python C2 agent. Upgrade to OpenSSH 10.6p2, restart sshd, AND hunt persistence.
- Chrome zero-day CVE-2026-125350: high-severity use-after-free in the XML component; the 8th actively exploited Chrome zero-day of 2026; fixed in 157.0.8103.50/.51 (Nov 9). Restart Chrome.
- CISA KEV Nov 9 wave (federal deadline Nov 12, now passed): GPUStack (unauthenticated admin access and secret disclosure), SignalWire FreeSWITCH (code injection via SIP MESSAGE), and IBM Aspera Faspex (authenticated user to full admin) — all actively exploited.
- N-able N-central "4EverAble2Shell" (active exploitation Nov 12): unauthenticated root RCE where the Legacy API is enabled and the platform is internet-reachable. Proof of concept early Nov 12, weaponized within the hour, 21 production servers mining Monero by 12:50 UTC. No fixed build — disable the Legacy API and remove internet exposure; treat as full RMM compromise.
- ScreenConnect / SCKit MSP campaign (Huntress, Nov 11): cloud-hosted panels abused from 17:47 UTC Nov 11, Run Commands installing rogue clients on at least 288 endpoints across 48 organizations in ~90 minutes. ConnectWise reports no compromise of its own infrastructure and no ScreenConnect vulnerability found; it disabled 5 extensions across ~400 partner environments. Scope remains under investigation.
- Microsoft Patch Tuesday (Nov 10) — compressed context: 1,259 total vulnerabilities per The Hacker News (Tenable/Ivanti count 1,093 Microsoft CVEs). Only one reported exploited: Outlook CVE-2026-124194, where opening a crafted .msg file connects to an attacker server and can disclose an NTLMv2 hash (KEV Nov 10, due Nov 24). Patch volume is background; the exploited item is the priority.
- D-Link DGS-3130 (unpatched exposure, no exploitation reported): multiple flaws in the D-Link Discovery Protocol and firmware security components of DGS-3130 switches; no CVEs and no fixed firmware yet. Mitigate: disable DDP, remove default SNMP community strings, restrict management interfaces to isolated networks.
Major confirmed incidents
- TSMC (confirmed Nov 12): some current/former employee data leaked, preliminarily matching records held by a third-party cloud provider; no evidence internal IT or fab systems were compromised; no production/customer data. Sellers claim ~46,000 records / 6.9 GB for 3 BTC — the total is NOT confirmed by TSMC.
- Maersk (confirmed Nov 11): unauthorized access to a local legacy system from a past acquisition; data extracted; system isolated; no wider/API/EDI impact; preliminary findings suggest no sensitive data disclosed. A group's claim of ~100 GB and a $5M demand is NOT confirmed.
- Baker Tilly (notices Nov 9): a third-party platform incident (no intrusion into Baker Tilly's own network found); Texas AG filing lists 10,456 residents (names/addresses/SSNs), South Carolina 1,359; no total disclosed; no group has claimed it.
- Vanderbilt University (confirmed Nov 9): active incident with shutdown/disconnect instruction; WiFi restored Monday evening and core services (email/Zoom/Brightspace/Okta/enrollment) available Tuesday evening, most other systems still restricted. A ransomware group's claims (>1M SSNs etc.) are NOT confirmed by Vanderbilt.
Offensive capability and adversary trends
- Autonomous/agentic offense is now documented at operational scale (Cairn, above), and Malbear Labs' reported "Operation Snowball" describes a jailbroken DeepSeek Flash framework used against a European telecom provider and 13 hosted companies via the Cisco BroadWorks flaw — reported research, not CISA-confirmed.
- Microsoft Research "Lazarus" (Nov 10, research only, NOT observed in the wild): editing fewer than 10 dormant neurons in an agent model plants a triggerable backdoor that emits an attacker command with the agent's permissions. Requires checkpoint write access. Defense: treat model weights like code and diff against a trusted base.
- ForgeRock automaker chain (Sam Curry / Shubham Shah, disclosed Nov 6, authorized research): WAF path confusion + Apache Felix console default credentials + OpenIDM scheduler RCE yielded sessions for any customer of two automakers, including connected-vehicle location/telemetry, lock/unlock, and remote start. Fixed within ~2 hours of report; no evidence of criminal use.
Defensive response and remediation developments
- Europol Operation Endgame (announced Nov 12): Gootloader, CastleRAT, and StealC networks disrupted — 235 servers and 257 domains seized across 9 countries, 6.4M victim datasets secured, 681 websites remediated.
- DOJ / Binance (Nov 12): court authorization to seize or restrain >$1.2B of the >$6B Binance theft (FBI TraderTraitor / North Korea), within >$2.7B frozen. Authorization is not completed forfeiture. The related FaceStealer deepfake video-call campaign (>$400M over ~18 months) was disrupted, with 900+ webcam videos seized. Binance's recovery continued late in the week, with withdrawals reopening across most networks.
ICHI score trends
No live numeric score is quoted: the live score did not render in the pre-production check, and inventing one would be worse than omitting it. Production methodology remains VECTORS v7.1 (seven scored dimensions; outages informational, not scored). The methodology page records a production revision dated November 12 (nvd-window-rates) fixing how weekly NVD vulnerability counts are read, after one batch of 21 exploit-tagged CVEs moved the published score by about 7 points with no change in the real world. Verified drivers this week: exploitation concentrated in remote access, identity, and management tooling with short federal deadlines; a sharp supply-chain pressure rise from the GitHub pipeline campaign; documented autonomous-AI actor operations; confirmed breaches concentrated in third-party and legacy systems. Model C refinements remain under development; no methodology change is announced.
Actionable observations
- Audit GitHub for "docs: dependency audit notes" PRs; rotate CI/cloud/AI secrets and review cloud trust policies and federated credentials — rotation alone does not evict this campaign.
- Treat RMM platforms (N-central, ScreenConnect) as tier zero: disable the N-central Legacy API, remove internet exposure, and review technician workstations.
- Verify the Barracuda ESG fix (9.4.0.030) actually landed; the KEV deadline was Nov 13.
- On musl/OpenSSH systems, upgrade + restart sshd and hunt persistence (authorized_keys changes, TAC Agent) — patching alone is insufficient.
- Restart Chrome for the zero-day fix; prioritize the single exploited Outlook flaw over Patch Tuesday volume.
- Inventory third-party platforms and legacy systems holding personal data — that is where this week's confirmed breaches lived.
Transcript
Akari: The weekly cyber rundown, brought to you by I C H I dot global. I'm Akari, an A I generated voice, and this is the week in review for the week ending Sunday, November fifteenth, twenty twenty-six. If last week was about the second door, this week is about the machine walking through it. The headline story is an autonomous A I framework running real intrusions almost end to end. Around it, the week's other big stories share the same shape: trusted systems, remote management tools, code pipelines, and partner platforms, turned against the people who rely on them.
Akari: We start with the A I breakthrough, because it changes the picture. Palo Alto Networks Unit forty-two reported this week that attackers used Cairn, an open source autonomous penetration testing framework, paired with two large language models, to run twenty-five operations against at least nineteen organizations between August and September. The researchers confirmed access or stolen data at ten of those organizations, across the semiconductor supply chain, defense policy, and civil society. At one target, more than one million member records were logged. Unit forty-two assesses with moderate confidence that this is a China linked campaign, and that it may have been run by one operator or a very small team. The entry points were not exotic. They were weak passwords and open, exposed endpoints. What changed is the labor. One small team, with an autonomous agent doing the reconnaissance, exploitation, and movement, reached a scale that used to take a much larger crew. That is the story of the week, and it is why the defensive lesson is unglamorous: the agent still needed a weak password or an open door to start.
Akari: The second machine scale story hit the code pipeline. Aikido Security reported a campaign they call w f x loot, which began on November tenth. Using stolen GitHub credentials, the attackers compromised four thousand eight hundred thirty-seven accounts and pushed a secret stealing payload into more than sixty-one thousand repositories through innocent looking pull requests titled docs, dependency audit notes. The payload ran in at least one thousand eight hundred six repositories, stealing pipeline secrets, cloud credentials, and A I tool keys, and in some cases rewriting cloud trust policies so the access survived a credential rotation. If your organization builds software: audit recent pull requests with that title, rotate pipeline secrets, and check cloud trust policies, not just passwords, because this campaign was designed to outlive a simple reset.
Akari: Remote management tools were the third front. Huntress reported that cloud hosted ScreenConnect panels were abused starting November eleventh, using the Run Commands feature to install rogue clients on at least two hundred eighty-eight endpoints across forty-eight organizations in about ninety minutes. The goal appeared to be the technician workstations behind those panels, the most privileged seats in a managed service provider. ConnectWise says it found no compromise of its own infrastructure and no vulnerability in ScreenConnect itself, and it disabled several extensions across about four hundred partner environments while the scope is investigated. Separately, an unauthenticated remote code flaw in N able N central, a remote monitoring platform, moved from proof of concept to live exploitation within hours on November twelfth. Twenty-one production servers were found mining Monero the same day. There is no fixed build yet. The vendor guidance is to disable the legacy A P I and get the platform off the open internet. Two different tools, one lesson: anything that can run commands everywhere must be treated as tier zero infrastructure.
Akari: Now the confirmed breaches. T S M C, the world's largest chipmaker, confirmed that data on some current and former employees was leaked, pointing to records held by a third party cloud provider. There is no evidence its internal systems or factories were compromised, and no production or customer data is involved. Sellers claim about forty-six thousand records. That total is their claim, not T S M C's confirmation. Maersk confirmed unauthorized access to a legacy system from a past acquisition. It was isolated, and preliminary findings suggest no sensitive data was disclosed. A ransomware group's claim of one hundred gigabytes and a five million dollar demand is not confirmed. Baker Tilly disclosed that a third party platform incident exposed client names, addresses, and Social Security numbers, with more than ten thousand Texas residents in the state filing alone. Vanderbilt University spent the week recovering from a confirmed incident that forced shutdowns, while a group's data claims remain unconfirmed. Late in the week, A X A Direct Japan said unauthorized access to its customer portal may have affected about five hundred ninety-seven thousand accounts, and Monex Securities said a third party account opening system may have exposed up to one point one three million email addresses and about sixteen hundred identity document images. Both are may have leaked maximums, and both say no passwords or card data were in the affected systems. Australian retailer Coles is also investigating an incident, with no evidence so far of compromised customer data. The pattern is third parties and legacy systems carrying the exposure, while the core systems, so far, held.
Akari: On exploitation, the week's confirmed list is short and serious. A two flaw chain against Barracuda Email Security Gateway appliances, a pre authentication bypass chained with a command injection, gives attackers root access. Barracuda says a small number of organizations were hit, a fix was pushed automatically, and the federal remediation deadline was November thirteenth. If you run one of these appliances, verify the fix actually landed. The Open S S H flaw on Alpine style Linux systems we covered last week moved from scanning to confirmed post exploitation this week, with attackers reading password hashes, modifying authorized keys, and deploying a Python based command and control agent. Patching is not enough there. Upgrade, restart the service, and hunt for persistence. A Chrome zero day, a use after free flaw in its X M L component, is the eighth actively exploited Chrome zero day of the year, and the fix is out, so restart your browser. And a wave of C I S A listings early in the week covered actively exploited flaws in G P U Stack, SignalWire FreeSWITCH, and I B M Aspera Faspex, all with a federal deadline of November twelfth that has now passed. Microsoft's Patch Tuesday, by contrast, was enormous in volume, more than a thousand vulnerabilities, but only one was reported exploited: an Outlook spoofing flaw, C V E twenty twenty-six, one two four one nine four, where opening a crafted message file can quietly connect to an attacker's server and leak an N T L M hash. Treat the volume as background. That one Outlook flaw, and the exploited list above, are the priority.
Akari: Two research stories deserve a careful mention, because neither is a confirmed criminal campaign. Researchers showed that basic mistakes in a ForgeRock identity deployment for two automakers, a firewall path confusion, a management console still using its documented default password, and a scheduler that could run code, would have let them create a session for any customer, including, for connected cars, location, door locks, and remote start. The automakers fixed it within about two hours of the report. And Microsoft Research showed a technique called Lazarus, where editing fewer than ten dormant neurons in an agent model plants a backdoor that fires only on a trigger. It has not been seen in the wild, but the defense is simple: treat model weights like code, and compare them against a trusted base.
Akari: The defensive bright spots were real. Europol's Operation Endgame took down three malware networks, Gootloader, Castle R A T, and StealC, seizing two hundred thirty-five servers and two hundred fifty-seven domains across nine countries, and securing six point four million victim datasets. In the United States, a court authorized the Justice Department to seize or restrain more than one point two billion dollars tied to the six billion dollar Binance theft attributed by the F B I to North Korea. Authorization is not completed forfeiture, but it is a large, concrete step.
Akari: So what does the I C H I index make of this week? I C H I, at I C H I dot global, measures the condition of the wider digital ecosystem on a zero to one hundred scale, where higher means worse, across seven scored dimensions under its production methodology, VECTORS version seven point one, with outages tracked for information but not scored. I am not quoting a live numeric score this week, because the live score did not render in our pre production check, and inventing one would be worse than omitting it. The verified drivers are clear without it. Exploitation pressure stayed concentrated in remote access, identity, and management tooling, with unusually short federal deadlines. Supply chain pressure rose sharply with the GitHub pipeline campaign. Actor operations now include a documented autonomous A I campaign at real scale. The methodology page also records a production revision dated November twelfth that fixed how weekly vulnerability counts are read, after one batch of disclosures moved the published score by about seven points with no change in the real world, a reminder that the band matters more than any single integer. Proposed scoring refinements, including the Model C work, remain under development, and no methodology change is announced here.
Akari: Your actions for the week ahead. One, audit your GitHub organization for the w f x loot pull requests, rotate pipeline secrets, and review cloud trust policies. Two, treat remote management platforms as tier zero: patch them, disable legacy A P Is, and keep them off the open internet. Three, verify the Barracuda fix landed, and on Alpine style systems, patch Open S S H and hunt for persistence rather than assuming the patch was enough. Four, restart Chrome to pick up the zero day fix. Five, inventory the third parties and legacy systems that hold your people's data, because that is where this week's confirmed breaches lived. I'm Akari. Stay safe, stay frosty, and I will see you next week.
Sources
- Palo Alto Networks Unit 42 — Attackers Use Autonomous AI to Target Semiconductor, Defense and Civil Society Sectors (Cairn) unit42.paloaltonetworks.com
- Help Net Security — Attackers breached 10 organizations with an open-source AI pentest framework www.helpnetsecurity.com
- Aikido Security — wfx-loot campaign pushes CI secret stealer to tens of thousands of repositories www.aikido.dev
- Help Net Security — Check your GitHub repos: A secret stealer hit tens of thousands of them www.helpnetsecurity.com
- The Hacker News — ScreenConnect Attacks Target at Least 288 Endpoints Across 48 Organizations With Rogue Clients thehackernews.com
- Huntback — 4EverAble2Shell exploited: Monero miners on live N-central huntback.io
- GBHackers — Europol Dismantles Three Major Malware Networks, Seizes 235 Servers and 6.4 Million Stolen Records gbhackers.com
- CoinDesk — DOJ gets court approval to seize $1.2 billion tied to North Korea's Binance hack www.coindesk.com
- Crypto Briefing — TSMC confirms employee data leak as hackers seek 3 BTC for the records cryptobriefing.com
- Neowin — Shipping company Maersk suffers data breach, ransomware gang demands $5 million www.neowin.net
- SC Media — Baker Tilly data breach exposes client information www.scworld.com
- The Register — Ransomware threat blares from Vanderbilt University computers during cyberattack www.theregister.com
- Sunday BFSI Cyber Threat Intelligence Briefing — 15 November 2026 (AXA Direct Japan and Monex Securities company disclosures) intelguys.github.io
- Reuters — Coles probes cyber security incident, says no evidence customer data compromised www.reuters.com
- BleepingComputer — Hackers now exploit critical Barracuda ESG flaws to gain root access www.bleepingcomputer.com
- The Hacker News — Microsoft Patches 1,259 Flaws, Including Exploited Outlook Bug That May Expose NTLMv2 Hashes thehackernews.com
- The Hacker News — CISA Warns of Active Exploitation of OpenSSH, GPUStack, FreeSWITCH, and IBM Flaws thehackernews.com
- The Hacker News — Google Issues Chrome Security Update for Eighth Actively Exploited Zero-Day of 2026 thehackernews.com
- BleepingComputer — D-Link warns of multiple zero-days in DGS-3130 Series switches www.bleepingcomputer.com
- Sam Curry — Hacking Two Automakers: ForgeRock Path Confusion, RCE, and Remote Vehicle Access samcurry.net
- Cloud Security Alliance Labs — Seven Neurons to Root: Lazarus Weight-Edit Backdoors in Agent Models labs.cloudsecurityalliance.org
- CISA — Known Exploited Vulnerabilities Catalog www.cisa.gov
- ICHI.global — Internet Cyber Health Index Methodology (VECTORS v7.1) ichi.global
Get ICHI Daily and ICHI Weekly by email
- ICHI Daily: one email on weekdays, after each new daily episode is published (no episode, no email).
- ICHI Weekly: one email on Sundays after 9 p.m. US Eastern time, once the new edition is published. If you choose both the weekly podcast and the written brief, they arrive together in that one email.
No new episode, no email. Every email links to the episode on ICHI; audio is never attached.
All ICHI Weekly episodes · ICHI Briefs & Podcasts
ICHI Daily and ICHI Weekly are editorial commentary on ICHI’s evidence. They never change the VECTORS score, incident evidence classifications or intelligence rankings.
ICHI Briefs & Podcasts · ICHI Daily · ICHI Weekly · Episode calendar · ICHI Perspectives · episodes.json · ICHI dashboard · Methodology