Binance confirmed this week what last week’s edito could only hedge on. Attackers stole $6.43 billion, up from the “up to $5.8 billion” the exchange was still investigating seven days ago. The how is the interesting part. Nobody broke the vault. The attackers ran fake job interviews, got malicious software onto employees’ computers, and then abused an internal AI tool and wallet system to approve unauthorized transactions. Ars Technica reports they spent months planning it.
Last week it was the freeze button that limited the damage. This week we learn what made the freeze necessary. Cryptography, cold storage and custody architecture were not the failure. The failure sat in two places most threat models file under someone else’s department: the hiring pipeline, where a request to run a coding test or install a meeting app looks like any other Tuesday, and the internal tooling trusted to approve what it was shown. An AI tool with authority over transactions is a privileged identity, and the attackers treated it like one. The security budget protected the vault while the way in was a calendar invite.
Table of Contents
🔓 BREACHES & SECURITY INCIDENTS
🇰🇵 Binance says attackers stole $6.43 billion after using fake job interviews and malicious software to compromise employees’ computers. They also abused an internal AI tool and wallet system to approve unauthorized transactions. Binance rebuilt affected systems, strengthened security, and is working to trace and freeze the stolen assets.
🔙 🚪 Hackers breached cPGuard’s update system and used it to install a backdoor and steal sensitive data from nearly 3,000 servers. The affected versions were 5.88.12 through 5.88.19, and stolen information included passwords, database settings, and SSH files. Admins should update to version 5.88.25 or later and change exposed passwords, keys, and other credentials.
🇬🇧 🏫 A ransomware group attacked a UK school network and encrypted about 37 computers. Researchers suspect the attackers may have used AI to automate some commands, but they could not confirm it. The attack also stole credentials and data, spread across school systems, and tried to disable security tools and backups.
🇺🇸 Hackers stole personal data linked to about 851,000 Harvard accounts — The leaked information included contact details and donation records, but not passwords or payment data. The ShinyHunters gang claimed responsibility and leaked the data after Harvard refused to pay a ransom.
🇺🇸 Marathon Health is notifying nearly 5.7 million people after a hacker stole data in an August scam. Stolen information may include names, birth dates, contact details, and Social Security numbers. The incident shows how criminals use impersonation to target companies holding large amounts of sensitive data.
🇪🇺 🧊 Hackers changed supermarket freezer settings across several European countries, spoiling food and causing major losses. Similar attacks in France show that insecure, internet-connected refrigerators can be remotely controlled. The incidents highlight how weak cybersecurity can disrupt food supplies and other essential services.
🇺🇸 Chainalysis says an AI-driven attack exposed some users’ names, email addresses, IDs, and work roles. The company says sensitive investigation data and passwords were not accessed, and it is still investigating. It advises users to watch for suspicious login or reset requests and enable multi-factor authentication.
🇬🇧 Northflank suffered a security breach that may have exposed customers’ stored secrets. Some customers reported unauthorized use of their AI API keys. Northflank urged customers to rotate credentials and check for unusual activity and charges.
💸 Polymarket will pay traders $3.9 million after a malicious script led to unauthorized trades and withdrawals. The script was linked to DebugBear, which says attackers hijacked its DNS records through its domain registrar. Both companies are investigating, and DebugBear says it found no evidence its internal systems were breached.
→ More incidents:
🥷🏻 CYBERCRIME, CYBER ESPIONAGE, APT’s
🥸 Three people are accused of using AI to impersonate executives and lawyers and steal about $129 million from companies. They allegedly tricked employees into wiring money, then converted it to cryptocurrency. They face fraud, identity theft, and money laundering charges.
🇺🇸 ⚖ 🇨🇦 Two Canadian men are accused of hacking US universities from 2016 to 2019 — They allegedly stole data and passwords, installed cryptocurrency-mining malware, and damaged computers. If convicted, each could face up to 10 years in prison and a $250,000 fine.
🇷🇺 Russian police arrested three people suspected of running the VasyGrek hacking group and seized equipment used in its crimes. The group used phishing emails and malware to steal money and target organizations across several countries and industries. Security researchers say VasyGrek’s attacks have grown more widespread and sophisticated.
🇷🇺 🇺🇦 Gorbag, a Russia-aligned group, targets Ukrainian government, law enforcement, and defense organizations. It uses phishing emails and a WinRAR flaw to install malware that steals browser data, VPN settings, and documents. The stolen information may support Russian intelligence efforts or help other attackers break into systems.
🐀 🐧 Hackers targeted government Linux servers to steal data and keep long-term access. They used a backdoor called SysGoRAT and scripts that may have been created with AI.
🇨🇳 A China-backed group used AI tools to attack over 200 organizations in 14 countries, compromising about 140 in 11 days. The targets included government agencies and companies in sensitive industries. AI let the group launch attacks much faster, and experts warn these campaigns are becoming more common.
🗓️ {Cyber,Info}Sec Events — A community-maintained list of infosec conferences worldwide. Subscribe to the ICS calendar feed to get events straight into your calendar, or follow @[email protected] on Mastodon for weekly digests. Contributions and ⭐ welcome!
👨🏻⚖️ 👀 GOVERNMENT, POLITICS, AND PRIVACY
🗾 Japan announced a $63 million plan to respond to a wave of cyberattacks — The plan uses AI and expert teams to detect threats and help affected organizations. It also aims to improve cooperation between government and businesses.
🇺🇸 🛜 Researchers found six White House network addresses being used by residential proxy services. The activity likely came from infected Windows 11 computers used in the evenings. Residential proxies have also been detected in remote places around the world.
🇺🇸 Democrats’ control of Congress could bring new attention and resources to cybersecurity agency CISA, AI risks, and cyber defense. Lawmakers may also try to renew a key information-sharing law before it expires in December. They are expected to review past cuts and changes to federal cybersecurity programs.
📲 🇩🇪 Germany warned lawmakers and public figures that spyware may have targeted their iPhones — Apple sent alerts to users in 99 countries, but the attacker and any successful hacks remain unknown. Officials advise anyone who receives an alert to assume their device may be compromised.
🦠 MALWARE & THREATS
🔙 🚪 🐧 Researchers found a Linux backdoor in Control Web Panel’s web server. It lets attackers run commands without leaving access-log entries, but installing it requires root access. Infected server owners should restore or update the server, change passwords, and check for other threats.
🎮 Hackers changed NIKKE’s PC launcher files to spread malware that could steal passwords, files, and cryptocurrency data. The publisher released a fix, but the full impact is still under investigation. Players should scan their computers and change sensitive passwords from a clean device if they may be affected.
🇷🇺 📲 APT29-linked hackers used compromised websites and a hijacked analytics script to target iPhone and Android users. Their exploits could steal cookies, Wi-Fi passwords, and other personal data from vulnerable devices. The report also warns that mobile exploit kits are spreading and being reused by more attackers.
ℹ VOLTA is Windows malware that steals passwords, financial details, and other personal data. It uses AI models to rank victims by how much their stolen information may be worth. The findings show how attackers are using AI to make stolen data easier to exploit.
🔐 💰 Synology warns that C3RB3R ransomware is attacking NAS devices running outdated software. Users should update DSM immediately, enable automatic updates, and block internet access to the device’s management page. The attacks encrypt files and demand thousands of dollars for a decryption tool.
🌏 Researchers found NetamLoader, a new malware framework targeting a government-only Windows 10 system in Asia. It uses trusted software components to hide its code and connect to a command-and-control server. The attack’s source and full impact remain unknown, though researchers found one victim.
🤖 🧰 AI, CRYPTO, TECH & TOOLS
🤔 OpenAI released a new cybersecurity model. Is that enough? — An OpenAI model helped a hobbyist find a serious Ubuntu flaw, which developers patched before any known attack. OpenAI is offering powerful cybersecurity tools to help defenders find and fix more flaws. But access is limited, and defenders may lack the time and money to keep up with AI-assisted hacking.
🇹🇼 One person used 12 AI agents to attack hundreds of organizations in 11 countries, stealing sensitive data. The attacks exposed serious weaknesses in emergency services, banks, universities, and other systems. The agents made mistakes, but the campaign shows how AI can help one attacker move quickly and why basic security fixes matter.
🍎 👀 Apple will scan iCloud photos and files shared through “Anyone with the Link” starting with version 27.3 updates. Sharing this way turns off end-to-end encryption, even if Advanced Data Protection is enabled. Apple may delete harmful content, restrict accounts, or contact law enforcement.
🐧 OpenAI’s AI pipeline found and tested Linux vulnerabilities, producing 948 exploits that could give local users root access in a 30-hour run. It also generated patches, but researchers reviewed the results before reporting them. The work shows AI can speed up both attacks and defenses, making fast, widespread patching essential.
⚔ NATO’s Crossed Swords exercise in Estonia trains teams from over 40 countries to coordinate cyber operations with military missions. This year, participants are also testing AI agents while keeping humans in control. Civilian organizations help teams practice protecting critical infrastructure.
🐛 🧠 VULNERABILITIES, RESEARCH, AND THREAT INTELLIGENCE
➝ From the Patching Department:
⚠ 🐛 D-Link found several unpatched security flaws in its DGS-3130 switches — Attackers on affected networks could steal device data, change settings, or disrupt service. Until patches are ready, admins should disable DDP and limit access to management services.
⚠ 🐛 A flaw in OpenSSH can let attackers access password-disabled accounts on some musl-based Linux systems by submitting an oversized password. It does not affect systems using glibc or PAM, and only works under specific settings. Administrators should update to OpenSSH 10.6p2 or disable SSH password authentication.
⚠ 🐛 A critical Rails flaw could let attackers upload a crafted image to write server files and, in some cases, read sensitive files. Apps are at risk if they accept untrusted uploads, use ImageMagick with Active Storage, and lack protective ImageMagick rules. Upgrade to Rails 8.1.4.1 or 8.0.5.2, and replace any secrets that may have been exposed.
⚠ 🐛 SonicWall found two security flaws in its end-of-life GMS software — It will not release fixes for the flaws. Customers should upgrade to SonicWall’s newer NSM platform.
🐛 💰Meta has paid about $1.2 million for bugs found in its Muse AI agent — The most serious flaw exposed data on internal test systems and earned a $300,000 reward. Meta says it fixed the reported issues and strengthened Muse’s defenses.
⚠ Barracuda warned that two flaws could let attackers gain root access to affected devices. The company released patches and urged customers to update firmware and restrict access to device admin pages.
🐛 💰 Google is offering a $250,000 reward for flaws that could give attackers full control of Google Cloud projects or organizations. Researchers must prove the risk without targeting real customer projects or using stolen access. Google also requires them to stop immediately if they achieve code execution in its production systems.
🛰️ ICS, OT & IoT
🤖 AI models can now help carry out many steps of attacks on industrial systems, especially with human guidance. This makes OT hacking easier for less-skilled attackers, though attacks can be noisy and defenses still slow them down. Experts say critical infrastructure faces added risk, but the threat should not be overstated.
🇬🇧 The UK’s cyber chief warns that AI-powered attacks on critical infrastructure are likely by 2028. AI may help criminals and hostile states find and exploit weaknesses faster, putting essential services at risk. He urges organizations to update technology and strengthen basic cybersecurity.
🐀 📲 Google disrupted a botnet infecting more than 1.3 million Android devices, mostly TV boxes. It says maker Videostrong deliberately built in a backdoor that criminals used for ad fraud and to route traffic through people’s homes. The backdoor remains, so owners may need to replace infected devices.
💥 Attackers are exploiting a critical Siemens IoT2050 flaw to run commands remotely without logging in. They have used it to install malware and create backdoors for future access. Siemens urges users to update to version 4.3.4.1 or later, or secure or remove Node-RED.
💬 CONNECT
Follow me on Mastodon for quick daily updates and bite-sized content.
Prefer using an RSS feed? Add Infosec MASHUP to your feed here.
Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.
Thanks for reading today’s newsletter, and if you're enjoying it and want to support my work, you can buy me a coffee ☕ over at https://www.buymeacoffee.com/0x58
See you next time!
-X.
