dogesec Logo
Obstracts Weekly Update #1: How a Fake Job Interview Became a $6.3bn Theft

Obstracts Weekly Update #1: How a Fake Job Interview Became a $6.3bn Theft

Binance filled 45 of 779 incident reports this week. Plus APT29's poisoned analytics script, ARTEX-run intrusions, Ivanti zero-days and the XCPE botnet.

Category: Research Series: Obstracts Weekly Update

Post series

Obstracts Weekly Update

The week's highest-scoring threat stories from the security blogs and feeds Obstracts reads, with links to the primary research.

Part 1 of 2

Part 2: Obstracts Weekly Update #2: US Seizes $1.2bn From North Korea and Dismantles Its Deepfake Meeting Scam →

tl;dr

  • Attackers attributed to North Korea’s TraderTraitor took about $6.34 billion from Binance’s hot wallets on 29 October; about $2.85 billion has been frozen.
  • Binance appeared in 45 of the 779 incident reports Obstracts scored between 2 and 9 November 2026.
  • 189 of those reports scored 80 or higher out of 100. The five below stood out.

How a fake job interview became a $6.3bn theft

Threat score 95/100 · Vectra AI, with ReversingLabs

The first reports on Binance covered the money. The ones that landed this week explained how the attackers got to it.

  1. Attackers attributed by the FBI to TraderTraitor, a North Korean hacking unit, approached developers with fake job interviews and a tampered download.
  2. From the compromised developer machines they moved between teams through an internal AI collaboration tool.
  3. They stole a container deployment token and used it to deploy a malicious container.
  4. That container signed withdrawals through Binance’s own legitimate signing paths, and roughly $6.34 billion left the hot wallets.

Vectra AI frames it as an ordinary enterprise breach that slipped through the gaps between endpoint, identity and network controls that were each working. ReversingLabs adds the toolkit it links to TraderTraitor: a trojanised Terraform provider (a plugin that lets the Terraform infrastructure tool manage a service) called FLATROOF, loaders hidden in WOFF font containers, and the ROOFDECK backdoor, with IoCs and YARA and Sigma rules for hunting. ReversingLabs is careful to say it cannot tie its own samples directly to the Binance intrusion.

If you want the full attack chain mapped out, The $6.34 Billion Job Interview: How North Korea Robbed Binance walks through what Binance, ReversingLabs and Vectra each saw, step by step.

APT29 poisons a web analytics script to hit iOS and Android

90/100 · The Hacker News

Sekoia attributes a watering-hole campaign it calls Poisoned Source to APT29. The operators injected malicious JavaScript into legitimate and lookalike sites, including a compromised Plerdy SEO Tracker script, fingerprinted visitors, and served N-day exploit chains (DarkSword, Coruna and AndroSnitch) to the ones worth targeting. The payloads take browser cookies, device data, Chrome data and Wi-Fi passwords.

Chinese APT uses AI agents to attack 200+ organisations

90/100 · Bleeping Computer

Vapor Panda, a China-linked group, used ARTEX, a multi-agent AI penetration-testing framework, to run simultaneous intrusions against more than 200 organisations in 14 countries. At least nine were compromised, across government, telecoms, defence, finance and nuclear. The agents handled password spraying, zero-day exploitation, web shells and the BLOODALCHEMY backdoor, and one chain went from first contact to root in 12 minutes.

Two more Ivanti Connect Secure zero-days, chained for root

90/100 · Bleeping Computer

Ivanti patched Connect Secure gateway flaws, including CVE-2026-104935, CVE-2026-111357 and CVE-2026-118527, that attackers have chained since early October for unauthenticated root access and file upload. Volexity, Defused and CISA have confirmed exploitation, and Shadowserver still counts thousands of exposed gateways.

Google disrupts a 1.3M-device Android TV box botnet

88/100 · Bleeping Computer

The XCPE backdoor shipped preinstalled in the firmware of Android TV boxes made by Videostrong, which investigators name as the operator with high confidence. It ran ad fraud and residential proxy services over MQTT command and control. Because it lives in the device partition, cleaning an infected box means reflashing it with clean firmware.

How this list is made

Obstracts reads security blogs and feeds and turns each post into structured threat intelligence. Every post that describes an incident gets a threat score out of 100 and a set of classifications (APT group, vulnerability, malware and so on). Each week we take every incident post published that week, group the ones covering the same story, and pick the highest-scoring stories with a mix of types. Where the research behind a story was published by the people who did it, we link to that rather than to a news rewrite.

Try Obstracts free for 7 days →

In summary

Binance was the week: a theft that began with a job interview and ended with legitimate signing paths approving the withdrawals. Around it, APT29 went back to poisoned third-party scripts, AI agents ran a Chinese espionage campaign at scale, Ivanti gateways needed patching again, and Google took down a botnet that shipped inside the hardware.

Posted by:

Dylan Thomas

Dylan Thomas, Do Only Good Everyday

Get new posts by email

New threat research and tutorials from dogesec, plus the weekly Obstracts threat brief. No spam, unsubscribe any time.

Thanks. Check your inbox to confirm your subscription.