Security
Binance loses $6.36 billion in North Korea-linked crypto heist
The largest crypto theft on record, a trojanized Claude Code extension, and an AI model that chained zero-days into working exploits.
This edition was produced with artificial intelligence. Text and voice are generated automatically.
Binance hit by $6.36 billion theft attributed to North Korea
On October 29, 2026, an attacker moved roughly USD 6.36 billion out of Binance wallets across ten blockchains, most of it within 20 minutes, making it the largest crypto theft on record — about four times the Bybit hack of February 2025. TRM Labs attributed the attack to TraderTraitor, North Korea’s hacking unit, with high confidence. On Ethereum, the attacker used an EIP-7702 delegation to turn a Binance hot wallet into a sweeper it controlled, draining 167,464 ETH and 63 tokens. Tether, Circle, Ripple and World Liberty Financial froze about USD 2.67 billion in stablecoins, and Arbitrum’s Security Council froze roughly USD 30 million more. Binance suspended withdrawals on several networks and said user funds are safe; it has not published a root cause.
North Korea Steals USD 6.36 Billion From Binance in the Largest Hack in History | TRM Labs →
Topics: Crypto securityState-sponsored hacking
Trojanized Claude Code extension shipped persistent SSH backdoor via Cursor Marketplace
A trojanized version of Anthropic’s official Claude Code extension was distributed through the Cursor Marketplace as a legitimate update, attempting to install a persistent SSH backdoor. The malicious version 2.1.313 was published under Anthropic’s real extension ID, anthropic.claude-code. Cursor said at most 5,000 users installed it between about 2:27 a.m. and 1:35 p.m. PT on October 29, 2026, and temporarily shut down its marketplace to remove the package. A 66-line loader added an attacker-controlled key labeled svc-telemetry to ~/.ssh/authorized_keys, set up a reverse tunnel to 134.209.82.225 and sent host reconnaissance data over unencrypted HTTP. Anthropic said its own systems were not compromised and that VS Code and OpenVSX versions are unaffected.
Official Claude Code Extension on Cursor Marketplace Trojanized with Persistent SSH Backdoor →
Topics: Claude CodeSupply chain attacks
Anthropic AI model chains flaws into working Chrome, Safari, Windows and Linux exploits
An unreleased Anthropic AI model found security flaws in Chrome, Safari, Android, Windows and Linux and chained them into working proof-of-concept exploits, Anthropic reported. The exploits were built in the lab after prompting and were not seen in a real attack; every flaw was reported to the vendor first and patched. Anthropic’s Frontier Red Team described five chains, including a Chrome exploit using four flaws in sequence and a Linux chain that yielded a root shell in 48 minutes. Anthropic says it reported 62 flaws to Google that were assigned CVEs and fixed as of Chrome 156, all rated critical or high. How much of each chain the model built on its own is not independently verifiable.
Topics: AnthropicAI-discovered vulnerabilities and AI bug-report slop
Volt Typhoon rebuilt attack infrastructure in under two minutes, Lumen finds
Volt Typhoon, a China-government-backed hacking group, has built and dismantled ephemeral malicious infrastructure to access hundreds of sensitive systems undetected, Lumen Technologies’ Black Lotus Labs reported. The operation, called Silent Train, is a series of compromised small office and home routers, IoT devices and enterprise infrastructure rotated rapidly since late 2023. Lead researcher Michael Horka said the group demonstrated a repeatable capability to rebuild a new chain with different infrastructure in less than two minutes, with sessions averaging roughly four hours. Black Lotus Labs has notified hundreds of victims, including telecom providers, government entities, defense contractors, universities and organizations in water, energy, transportation and emergency services. The group exploited more than 27 device classes, including products from Fortinet, Cisco, Ivanti, F5 and Palo Alto Networks.
Researchers uncover Volt Typhoon’s covert, ephemeral network | CyberScoop →
Topics: State-sponsored hacking
Gitee scan finds 26,400 live credentials across 7.25 million repositories
An independent security researcher scanned 7,256,531 public repositories on Gitee, China’s largest indigenous Git hosting platform, using TruffleHog and Kingfisher concurrently and found 26,400 unique verified-live credentials after deduplicating by token value. The scan ran on Modal at 350 concurrent containers over about one month, costing USD 3,097.64. Alibaba Cloud AccessKey pairs were the most common leaked credential type, at roughly 1 in every 766 repositories. 2025 alone accounts for 12,242 unique credentials, nearly half the dataset. Findings included an active organization-admin credential for a major enterprise AI platform and thousands of live cloud provider access keys. Gitee’s density of 3.64 verified-live credentials per 1,000 repositories exceeds Bitbucket and GitLab, though those scans used TruffleHog alone. Vendors with the highest exposure received coordinated reports; revocation status had not been confirmed.
Scanning 7.25 Million Public Repositories on China's Largest Git Platform for Secrets | MongoDB →
PDF tools advertise AES-256 but derive keys from tiny state spaces
Several PDF implementations advertise AES-256 encryption but generate the 256-bit file key from state spaces small enough to enumerate, enabling passwordless decryption of protected PDFs. In MuPDF versions before 1.28.5, only nine bits of random state affect the actual AES-256 key, so all 512 possible keys can be brute-forced; an unoptimized pure-Python search takes about 11 milliseconds. Foxit PDF Reader 2026.1.3.36551 derives the file key as SHA-256 of the current timestamp, 32 uninitialized bytes and the constant there, letting a Rust implementation search 370 to 376 million timestamps per second. OpenPDF 3.0.5 and iText 9.7.1 route randomness through an ARC4 stream seeded with a millisecond timestamp and JVM free memory. The issues were disclosed to vendors and fixed in MuPDF’s case; previously published copies cannot be retroactively protected.
The case of the 9 bit AES key | V12 →
Topics: Encryption
Linux kernel CVE-2026-98432 draws scores from 5.9 to 9.8
A use-after-free in the Linux kernel’s IPv4 TCP subsystem, published October 15, 2026 as CVE-2026-98432, has produced divergent risk scores. The Linux Kernel CVE Team assigns 9.8 out of 10, critical; Red Hat reports a range of 5.9 to 8.1, with its machine-readable advisory at 8.1 and its CVE page at 5.9; Amazon Linux scores it 7.0 for AL2023. The flaw resides in tcp_v4_syn_recv_sock() and has existed since kernel 4.4. The upstream fix, applied October 5, 2026, replaces direct kfree() calls with kfree_rcu(). As of October 29, Debian, Oracle Linux, SUSE and Amazon Linux have shipped corrective packages, while Ubuntu lists status work in progress and Red Hat Enterprise Linux shows fix deferred. No vendor advisory or CISA KEV listing reports in-the-wild exploitation.
CVE-2026-98432: The CVSS War Over the Linux Kernel TCP Vulnerability | DeafNews →
Topics: Zero-day vulnerabilities
TightVNC Viewer 2.8.88 carries pre-auth RCE via framebuffer integer overflow
TightVNC Viewer 2.8.88 x64 contains a pre-authentication remote code execution flaw in how it sizes its framebuffer. A server announcing 32768 x 32769 at 32bpp makes the buffer size wrap from 4 GB to 128 KB, while the viewer keeps the announced dimensions as its logical canvas, giving a server-controlled write at any signed offset. The root cause is in rfb/FrameBuffer.cpp, where getBufferSize() computes the product in 64 bits then discards it with an int cast. In listening mode, tvnviewer.exe -listen binds 0.0.0.0:5500 with no IP restriction and no password, putting the RCE directly on the network. The exploit chain leaks pointers, places a fake object and ROP chain in the framebuffer and calls CreateProcessW. TightVNC patched the vulnerability on October 3, 2026 without mentioning any security patch; the flaw was credited to SeongJoon Cho of SSD Labs Korea.
TightVNC Viewer Pre-auth RCE - SSD Secure Disclosure →
NIK C2 framework routes stolen data through OpenAI API from infected hosts
Cisco Talos observed an AI-integrated command-and-control framework in the wild called NIK C2. Its Windows implant lets an actor dump credentials, stream keystrokes and instruct an integrated AI model to analyze harvested data within the compromised session. The framework abuses the OpenAI Chat Completions API; if no key is supplied, it reads the OPENAI_API_KEY environment variable, using the victim’s own billing account. The Go-based implant uses TLS-secured WebSockets on port 443 plus an HTTP fallback polling loop every 250 milliseconds, with traffic protected by AES-256-GCM. It supports over 52 commands, including browser and cryptocurrency wallet theft, keylogging and a screen-lock ransomware module that encrypts no files. Talos assessed with moderate confidence that NIK C2 is a fully functional framework, not a proof of concept.
NIK C2: The new AI-enabled command-and-control framework →
Topics: AI-powered malwareOpenAI
VOLTA-AUTODOXER malware asks Claude to rank victims’ assets by value
Cisco’s Talos research organization reports a self-deleting Windows infostealer, VOLTA-AUTODOXER, that steals data from a victim’s PC and sends it to an AI model — Anthropic’s Claude by default — to produce a target assessment. The malware builds a structured evidence bundle on the person and asks the model for a list of their valuables ranked by how much they could be cashed out for. VOLTA prompts the model to act as a forensic intelligence analyst and to be blunt, specific and accurate. It can switch models if one refuses, or skip the AI step entirely. VOLTA steals browser passwords, cookies and saved payment cards, email profiles, Microsoft OAuth tokens, screenshots and SSH and cloud-provider configuration files, and attempts to crack cryptocurrency wallet vaults inline.
Topics: AI-powered malwareClaude
Hacker targeted high-stakes online poker players with card-revealing malware
Illicit software installed on the computers of some of the world’s best online poker players let an accused cheater see the cards in victims’ hands and beat them at the tables, WIRED reported. The hacker’s plans were foiled by an AI agent. Further detail on the malware, the players targeted and the AI agent’s role remains behind the paywall.
A Hacker Targeted High-Stakes Poker Players With Malware That Exposed Their Cards | WIRED →
Topics: Malware & infostealers