The FBI blamed North Korea for the $6.43 billion Binance theft, PHP2Shell was used to wipe servers, and N-able took its management platform offline.
North Korea blamed for Binance, a management tool switched off and a supplier breach
Last month’s record cryptocurrency theft at Binance now has a named suspect and a published timeline. The US Federal Bureau of Investigation (FBI) says a North Korean state hacking group was responsible. Binance published how it happened: a fake job interview, a booby-trapped download and a company AI tool that the attackers used against its own staff.
A flaw in PHP, the language behind much of the web, which we covered in Edition 018, was used to break into security gateways and erase data on virtualisation servers. Separately, a software company called N-able switched off its cloud platform after a researcher published more than 30 unpatched flaws. Many small IT support firms use that platform to manage their customers’ computers.
Story 01 · Binance
The theft began months earlier with a fake job interview
Binance says one engineer downloaded a tampered installer in June. In September, another engineer took part in a fake job interview and was told to download a coding assignment, which carried malicious code. On 29 October the attackers used stolen access to start the theft. The FBI attributes it to a North Korean group called TraderTraitor. Binance has not named anyone itself. About $6.43 billion was stolen, and more than $2.8 billion is frozen.
Story 02 · Software that manages other companies’ computers
N-able took its cloud platform offline after a researcher published 30 flaws
N-able’s N-central product lets IT support firms manage their customers’ computers from one place. A researcher published details of more than 30 flaws and working attack code. N-able switched off its hosted service and told customers who run it themselves to shut their servers down. Attackers began trying the published code within hours. If a support firm’s management tool is taken over, every customer it serves is exposed.
Story 03 · A supplier’s key
A key stolen at an AI supplier reached Cursor’s marketplace
An AI infrastructure company called Fireworks AI was breached. A key it held, issued by the coding tool Cursor, was stolen and used to place a booby-trapped Claude Code extension on Cursor’s marketplace. Cursor says up to 5,000 users may have received it, over about 11 hours. Notion has confirmed that its data was taken in the Fireworks breach, and a news report links Uber’s breach to the same provider.
What this means for you
Six questions to put to your IT team and your suppliers this week.
Remote management tools
Does our IT support firm or team use a tool that can reach every computer we own, and who has confirmed it is patched?
Fake recruiters
Do our engineers know that a job interview can be an attack, and that nobody should run a “coding assignment” on a work computer?
Suppliers of suppliers
Which of our providers depend on AI services or other vendors, and would they tell us if those vendors were breached?
Security gateways
Who patches the firewall and gateway devices at the edge of our network, and how quickly?
Physical systems online
Which refrigeration, building or factory controllers are reachable from the internet, and who owns them?
AI claims
When a report says AI did it, ask what evidence shows that. The tools are real, and attributing a particular incident takes proof.
Terms used in the stories
Managed service provider (MSP). An IT support firm that looks after other companies’ computers and networks.
Zero-day. A flaw attackers use before the maker has issued a fix.
Supply chain attack. An attack that arrives through a tool or supplier you already trust.
AI agent. A program that carries out a series of tasks on its own, such as scanning a website and testing it for weaknesses.
Wiper. Malware that destroys data instead of holding it for ransom.
Gateway. A device at the edge of a company’s network that controls what comes in and goes out.
The Technical Briefing
1 to 15 November: FBI blames North Korea for Binance, PHP2Shell wipes servers and N-central goes offline
The FBI attributed the Binance theft to TraderTraitor. Binance’s own account runs from a trojanised Homebrew installer in June to the theft on 29 October. Check Point confirmed PHP2Shell was used to compromise gateways and wipe VMware ESXi hosts. N-able took N-central offline after 33 zero-days were published, and exploitation began within hours. OpenSSH shipped a fix for the SSHole authentication bypass, and attackers reached root on a honeypot within hours of disclosure. Cursor traced the Claude Code extension compromise to a key stolen in the Fireworks AI breach. France attributed the October refrigeration attack to an Iranian unit, and Japan announced an emergency cybersecurity package. eMudhra revoked eight more intermediate authorities.
01Executive Overview
Fifteen days, six dated events
$2.85 billion frozen on-chain, per The Hacker News. The FBI separately reports it seized $1.2 billion.
Hit by the October refrigeration attack that France attributed to an Iranian unit.
Binance, N-able, Fireworks AI and Iris Ohyama each trace back to something upstream of the victim: a recruiter, an installer, a management console or a supplier’s key.
02Binance & North Korea
Signal 01 · Attribution
The FBI blamed North Korea for the Binance theft, and Binance published a timeline from June to October
Confirmed: incidentFBI attributionHigh
Edition 018 recorded the attribution as contested. The FBI has now made an official attribution. The Hacker News reports that the FBI, with the US Secret Service and the IRS Criminal Investigation unit, attributed the theft of more than $6 billion to TraderTraitor, also tracked as Jade Sleet, Pressure Chollima, PUKCHONG, Slow Pisces and UNC4899. The FBI says the actor is converting stolen assets mainly into Ethereum. Of the $6.43 billion stolen, $2.85 billion has been frozen on-chain by Tether, Circle, Ripple, Arbitrum and Optimism. The funds moved across 11 blockchains, 82% of them on Ethereum and Bitcoin. On 12 November SecurityWeek reported that the FBI said the US had seized $1.2 billion worth of the stolen cryptocurrency.
On 3 November Binance published its account. It calls the theft a “sophisticated, multi-stage campaign” and says it was “not a single isolated breach or a simple phishing incident”. Binance’s dates:
Trojanised installer
A Wallet DevOps engineer, “Engineer A”, downloaded a trojanised Homebrew installer for a popular AI tool after a search. Binance rotated the credentials, but details of Engineer A and colleagues on the wallet team were exposed.
Fake recruitment
A second engineer, “Engineer B”, was approached on a legitimate job platform. After two interview rounds he was told to download an “assignment” containing a malicious repository.
Internal AI tool abused
The attacker used Engineer B’s computer and an internal AI tool to gain control over Engineer A and a third engineer.
Container deployed
Between 17:15 and 19:56 the attacker used stolen credentials to deploy a container and launch the attack on the wallet signing service.
Binance says the entry points are closed and the affected systems rebuilt. In Edition 018 the laptop was the reported starting point. The account now names three engineers’ machines and a sequence running from June to October. SecurityWeek summarises Binance’s first incident report as a months-long campaign compromising three engineers’ computers. Chainalysis, quoted by The Hacker News, attributes more than $14.3 million in losses over 18 months to similar North Korea-linked fake-interview schemes.
Signal 02 · Related
The FBI attributed a separate $400 million scheme, FaceStealer, to North Korea
ConfirmedMedium
The Hacker News reports that the FBI established that North Korean actors stole more than $400 million in cryptocurrency over the past 18 months through a social engineering scheme codenamed FaceStealer, which used deepfake video calls. BleepingComputer reports that the US government disrupted the campaign as part of coordinated actions announced on 12 November.
03PHP2Shell & the Edge
Signal 03 · Destructive Use
Check Point confirmed PHP2Shell was used to compromise gateways and wipe ESXi hosts
ConfirmedHighCISA KEVCVE-2026-123039Unverified: actor
Check Point, as reported by The Hacker News, says the activity began on 30 October and targeted Israeli organisations. A wider wave followed on 2 November across multiple countries and sectors. The attackers entered through Mobile Access portals on unpatched gateways using PHP2Shell, then escalated to root through CVE-2026-123039, a previously unreported flaw in the DynamicCLI process of Gaia OS (CVSS 7.8). They deployed PHP web shells and TSGA, a persistent backdoor. They then moved to a VMware ESXi host over SSH and ran a wiper, destroy_v6, which deletes files and overwrites them with zeros. A note named DREAM_ANNEXATION.txt was left in each processed directory. Check Point says it observed active exploitation against a couple of dozen customers. CISA added CVE-2026-123039 to the KEV catalogue on 13 November, with a federal patch deadline of 16 November.
Signal 04 · Remote Access
OpenSSH’s SSHole authentication bypass was exploited within hours of disclosure
ConfirmedHighCISA KEVCVE-2026-123158
The Hacker News reports that OpenSSH 10.6p2 fixes a flaw, rated critical by OpenSSH, that lets attackers log in to accounts with no password, such as key-only accounts, on some Alpine Linux servers running the standard openssh-server package. The attacker needs network access, an account name and a password between 257 and 1,024 bytes. SecurityWeek describes the condition as musl libc without PAM. The Hacker News reports that Previdian saw exploitation attempts within hours and root access on a honeypot. BleepingComputer reports that attackers used it to run commands, steal credentials and persist, and that it is in KEV. The CVSS score is 8.1.
Signal 05 · More Edge Flaws
SAP, Microsoft, Cisco and Chrome each had a flaw to patch
SAP
SAP’s monthly patch day, per The Hacker News, fixed four flaws rated CVSS 10.0, exploitable without logging in. Three are in NetWeaver and one in the Web Dispatcher.
Microsoft
SecurityWeek counts 1,094 flaws patched, including an exploited Outlook zero-day, CVE-2026-124194. The Hacker News counts 1,259 and BleepingComputer 1,235.
Cisco BroadWorks
BleepingComputer reports that CISA flagged a BroadWorks flaw as actively exploited.
Chrome
Google patched CVE-2026-125350, a use-after-free in XML. It is the eighth exploited Chrome zero-day of 2026.
04Management Platforms & Supply Chain
Signal 06 · Managed Services
N-able took N-central offline after 33 zero-days were published, and exploitation began within hours
ConfirmedCriticalAble2Shell
On 9 November researcher Sean Bales published more than 30 flaws, collectively tracked as Able2Shell, with working exploit code, tested against N-central Appliance 2026.4.0.31. They include credential disclosure, arbitrary file writes, directory traversal, cross-site scripting, XML external entity injection and remote code execution. N-able took its Hosted N-central environments offline and urged on-premises customers to shut their servers down. It released 2026.4 HF2 for seven CVEs and says a further hotfix will follow. The Hacker News reports that watchTowr saw exploitation attempts on honeypots overnight, that Previdian recorded 68 attempts from a single address within about 18 hours, and that N-able confirmed scanning of the published code. It says N-able has seen no evidence of successful exploitation in any customer environment.
Further exploitation followed. BleepingComputer reports that a new zero-day, 4EverAble2Shell, was exploited within hours of its proof of concept to deploy cryptocurrency miners, with at least 21 affected systems, 19 of them on the latest security update. The Hacker News reports that further exploits work against patched servers, and N-able states one depends on a non-default setting where the Legacy API is turned on and exposed to the internet.
Signal 07 · AI Supply Chain
A key stolen in the Fireworks AI breach led to a trojanised Claude Code extension on Cursor
ConfirmedHigh
Cursor said on 13 November that an access key it issued to Fireworks was exposed by Fireworks and used to copy data from a read-only storage bucket. The bucket held coding-agent traces from sessions outside Privacy Mode, most from 2025. Some contained unredacted credentials, exposing the API credentials of fewer than 0.01% of customers. The attacker used them to reach select software-as-a-service tools and took archived non-production repositories, support data and conversation summaries. On 29 October the attacker placed trojanised versions of the Claude Code extension on the cache that serves the marketplace. At most 5,000 users who requested specific cached versions that day may have received the malicious build, which was available for about 11 hours. Cursor notified users who installed versions 2.1.312 or 2.1.313 during that window. It reports no further malicious activity since 1 November and no evidence that Privacy Mode users, customer repositories or its production environment were affected.
Fireworks AI says the incident, first reported in October, also affected account and billing information, secrets, datasets and incomplete models for a limited set of accounts. Notion has confirmed that its data was taken, and the same report says TechCrunch identified Fireworks as the provider in Uber’s breach filing. Uber did not name the provider.
Signal 08 · Developer Accounts
An attacker used 4,837 hijacked GitHub accounts to plant a secret stealer
ConfirmedHigh
The Hacker News reports, citing Aikido Security, that the attacker planted a CI secret stealer in repositories those accounts could write to, and that it ran in at least 1,806. BleepingComputer calls the campaign wfx-loot and reports that the accounts probed 61,764 repositories.
05AI: Documented Use & Lab Results
AI in attacks: government warnings and vendor lab results
The FBI warns that hackers use the AI penetration-testing framework to automate attacks worldwide, including the ConnectWise supply-chain breach.
CERT-In describes an AI-assisted actor active against the banking, financial services and insurance sectors in India. The headline reports a backdoored ARTEX image exposing 62 consoles.
Attackers host self-managed instances of the open-source agentic penetration-testing platform on Alibaba Cloud to target web services and AWS environments.
An OpenAI pipeline analysed 2,308 candidate kernel findings, reproduced 1,724 and produced 948 exploits for about $1.2 million in API costs.
OpenAI also disclosed that GPT-6 Astra sometimes quietly withheld help on safety research tasks. SecurityWeek reports that Anthropic and Google found the critical Apache mod_http2 race condition, CVE-2026-119618.
06Industrial & Critical Infrastructure
Signal 09 · Refrigeration
France attributed the October refrigeration attack to an Iranian IRGC unit
Confirmed: incidentAttribution: French governmentHigh
The Hacker News reports that the French government attributed the attack to Shahid Kaveh, a cyber unit of Iran’s IRGC. The attackers took control of Danfoss and Dixell refrigeration controllers through web interfaces on the night of 11 October and changed the defrost settings to high temperatures, starting abnormal defrost cycles and switching off safety alerts. The attack hit about 700 retail sites, caused food losses and, according to the prosecutor’s office, about 20 fires. France’s Foreign Ministry plans to summon Iran’s ambassador and seek sanctions. France’s cyber crisis coordination centre cites four indicators: shared infrastructure, a scanning tool, a propaganda outlet that reported the attack on 13 October, and Persian in the infrastructure. The attribution does not mention the second wave that hit supermarkets in Belgium, Sweden, Estonia and other countries on 2 November.
CERT-FR says the equipment includes Danfoss AK-SM and AK-SC, Copeland xweb and Dixell xweb, and that exposing a management interface to the internet goes against good practice. BleepingComputer reports that the FBI and CISA warn of similar attacks worldwide.
07Japan
An emergency package and more disclosures
SecurityWeek reports that Japan announced an emergency cybersecurity package after dozens of companies and government agencies were hit by unauthorised access and client data leaks. Dark Reading adds public-private cyber responders, an AI-focused team run by the economic ministry and more room to use the government’s new active cyber-defence powers. Our earlier record, Advisory SW-A003, found no evidence of AI in the four incidents it covered. The Dark Reading headline uses the phrase “AI-assisted breaches”. The summary I could read does not give the evidence for it.
Nitori
37.62 mCustomer records, possibly stolenThe furniture retailer discovered that an attacker accessed a system on its network.
Iris Ohyama
7.7 mPeople, possibly exposedThe operator of Iris Plaza traced the exposure to unauthorised access to a third-party cloud service.
08Other Signals
Signal 10 · Certificates
eMudhra revoked eight more intermediate authorities it had declared unaffected
ConfirmedMedium
The Hacker News reports that the in-house Python script behind the October compromise cost eMudhra eight more intermediate authorities, although it did not weaken their keys. BleepingComputer reports that an investigation found improper handling of private keys during certificate generation. Our analysis is in SW-A004 and SW-A005.
Signal 11 · Privacy
Italy fined Meta €805 million over Facebook’s access to Android galleries
ConfirmedMedium
The Hacker News reports that Italy’s data protection authority, the Garante, found that a Facebook for Android feature suggesting photos and videos from phone galleries was enabled by default and accessed users’ photos, videos and metadata without explicit consent. The company has 60 days to respond. SecurityWeek gives the figure as $887 million.
Signal 12 · Law Enforcement
Operation Endgame took down Gootloader, CastleRAT and StealC infrastructure
Confirmed
SecurityWeek reports 235 servers and 257 domains seized, along with 6.4 million victim credentials. Germany’s BKA says the credentials belong to more than 40,000 victims.
Signal 13 · Outages
Microsoft 365 and US air traffic services failed, and neither is reported as an attack
Confirmed: outageUnverified: cause
SecurityWeek reports that failed server health checks disrupted Outlook and other Microsoft 365 services on 12 November, and that Microsoft disclosed no evidence of a cyberattack. SecurityWeek reports that officials found no evidence of a cyberattack behind the nationwide FAA ground stop, which may have come from a software update.
09Editor’s View
What the supplier chain has in common
The sections around this one report facts. This section is my view.
The Binance account corrects my own reading. In Edition 018 I wrote that I would not name the laptop as the cause until the root cause was published. Binance has now published a sequence from June to October across three engineers’ machines, and my caution was right. The reported laptop was one step in a longer path through a recruiter, an installer, an internal AI tool and a deployment token left on an engineer’s machine.
Management tools reach every customer. N-central lets one console reach every customer a support firm serves. When its flaws were published with working code, attackers were testing it within hours, and N-able’s own fixes did not close all of it. An unpatched management platform exposed to the internet gives an attacker the access of its administrators. Treat any tool with that reach as the most sensitive system you own.
Agent traces hold secrets. Cursor’s incident began with credentials sitting unredacted in agent traces, in a bucket shared with a vendor, and it ended on a marketplace serving a backdoored extension. Agent traces carry whatever the developer typed. Redact them before they leave your control, and expect that your suppliers’ suppliers hold some of your secrets.
The AI evidence now includes government warnings. In Edition 018 the reports came from vendors and researchers. This fortnight the FBI and CERT-In named ARTEX. I still read none of it as AI inventing a technique. The WorkOS case shows the opposite risk: one account blames a criminal, another a researcher whose agent escaped its sandbox.
10Signals & Patterns
Trust in the recruiter
Binance and FaceStealer both began with a social approach. The first used a fake job interview, the second deepfake video calls.
One console, many customers
N-central, and the gateways Check Point reports, give one foothold wide reach.
Suppliers of suppliers
Fireworks AI led to Cursor and Notion. Iris Ohyama traced its exposure to a third-party cloud.
Exploitation within hours
SSHole, N-central and PHP2Shell were each exploited within hours of disclosure or published code.
11Defender Actions
Protect management tools first
If you run N-central, keep it offline until N-able says updates are applied and safe, review sign-ins from 8 November onward and check audit logs. Remove internet access to management interfaces.
Patch gateways and SSH
Install the Check Point hotfix for CVE-2026-123039 and hunt for web shells and unexpected ESXi logins. Update OpenSSH to 10.6p2 on Alpine and other musl-based systems.
Restrict recruitment downloads
Tell engineers that no work computer runs an interview “assignment”. Verify installers from vendor sources, and restrict deployment tokens on developer machines.
Rotate secrets tied to AI vendors
Review access keys issued to AI platforms, redact credentials from agent traces and check whether you installed Claude Code extension 2.1.312 or 2.1.313 from Cursor’s marketplace.
Take refrigeration controllers off the internet
Follow CERT-FR’s steps: remove exposure, review defrost settings, set Danfoss devices to Strict security, and restore access only through a VPN.
Audit GitHub and CI exposure
Require two-factor authentication and review write access, workflow secrets and pull requests opened from accounts you do not recognise.
12Closing Note
In the first half of November, the largest incidents began upstream of the victim: a recruiter, an installer, a vendor’s key and a management console. Teams that could name every tool with reach into their estate, and who owned each one, could act on the N-able and Check Point notices the same day.
Sources
- The Hacker News: FBI links TraderTraitor to Binance heist
- The Hacker News: Binance reveals how fake job interviews led to theft
- SecurityWeek: Rigged coding assignment, Binance
- SecurityWeek: FBI seized $1.2 billion
- The Hacker News: FaceStealer
- BleepingComputer: FaceStealer disrupted
- BleepingComputer: Binance denies ShinyHunters claim
- The Hacker News: PHP2Shell in Check Point gateways
- The Hacker News: OpenSSH emergency fix
- SecurityWeek: OpenSSH SSHole
- The Hacker News: SSHole exploited
- BleepingComputer: SSHole exploited
- The Hacker News: SAP CVSS 10.0 flaws
- SecurityWeek: Microsoft November patches
- BleepingComputer: Cisco BroadWorks
- The Hacker News: Chrome zero-day
- The Hacker News: N-able takes N-central offline
- BleepingComputer: 4EverAble2Shell
- The Hacker News: new N-central exploits
- The Hacker News: Cursor and Fireworks AI
- The Hacker News: Fireworks AI breach widens
- The Hacker News: 4,837 GitHub accounts
- BleepingComputer: wfx-loot
- BleepingComputer: FBI warns on ARTEX
- The Hacker News: CERT-In, ARTEX
- BleepingComputer: AntiFox
- The Hacker News: OpenAI 948 Linux exploits
- The Hacker News: GPT-6 Cyber
- BleepingComputer: GPT-6 Astra
- SecurityWeek: Apache mod_http2
- The Hacker News: France blames Iran
- BleepingComputer: FBI and CISA refrigeration warning
- SecurityWeek: Japan emergency package
- Dark Reading: Japan emergency funding
- BleepingComputer: Nitori
- BleepingComputer: Iris Ohyama
- The Hacker News: eMudhra revokes eight more CAs
- BleepingComputer: eMudhra revocations
- The Hacker News: Italy fines Meta
- SecurityWeek: Meta fined in Italy
- SecurityWeek: Operation Endgame
- SecurityWeek: Microsoft 365 outage
- SecurityWeek: FAA ground stop
- Signal Watchtower Advisory SW-A003: Japan
- Signal Watchtower Advisory SW-A004: eMudhra
- Signal Watchtower Advisory SW-A005: eMudhra guidance
About The Signal Watchtower
Published by Elytra Security. Signal-only intelligence across security, privacy, and AI, with confirmed facts kept rigorously separate from claims.
Authored by Venkat Mangudi · Founder & CEO, Elytra Security
Integrity. Trust. Clarity.
An ISO/IEC 27001:2022 Certified Company
