SecurityWeek’s weekly cybersecurity news roundup offers a concise overview of important developments that may not receive full standalone coverage yet remain relevant to the broader threat landscape.

This curated summary highlights key stories across vulnerability disclosures, emerging attack methods, policy updates, industry reports, and other noteworthy events to help readers stay well-informed about the evolving cybersecurity environment.

Here are this week’s highlights: 

ICS-Sniper attack picks off critical OT traffic

University of British Columbia researchers have detailed ICS-Sniper, a proof-of-concept attack aimed at industrial facilities that rely on cloud-hosted SCADA systems. An attacker who controls a router on the network path could study the size, direction and timing of VPN-encrypted packets to predict when critical commands or data will be exchanged, then drop payload-carrying packets during those windows. In tests on emulated versions of the SWaT water treatment testbed, the attack delayed the start of clean water pumping by about 15 minutes and, in a separate test, kept a valve open more than nine minutes longer than intended.

Intel pulls SGX from Xeon roadmap after Diamond Rapids

Advertisement. Scroll to continue reading.

Intel says Diamond Rapids, its next-generation Xeon platform, will be the last to include SGX, as it focuses its confidential computing roadmap on TDX. The company says the industry has moved toward VM isolation technologies like TDX because they’re easier to deploy and scale. Platforms with SGX will keep receiving support and security updates into the early 2030s, giving customers several years to migrate.

XWorm operators’ panels open to takeover 

Mootoo, an independent researcher based in Oman, has published a PoC exploit for an unauthenticated RCE flaw in the XWorm RAT’s C2 panel (versions up to 7.4). The researcher found it on October 4 while reverse-engineering an XWorm sample used in an attempt to compromise them. An attacker who has the panel’s address, port and communication key (all of which can be pulled from an XWorm client’s configuration) can pose as an infected machine, overwrite a DLL through path traversal, and get code running on the C2 host without any action from the operator. Mootoo told SecurityWeek they used the flaw to compromise infrastructure tied to the person behind the attack, and that it appears to be exploited on a limited scale.

Bank of America brings in patch army after Mythos

Bank of America CEO Brian Moynihan told CNBC that the bank brought in 1,000 outside contractors in the second quarter to help its 3,000-strong internal team fix flaws in proprietary and vendor software exposed by frontier AI models such as Anthropic’s Mythos. The contractors are still working there. Moynihan said the remediation effort, together with anti-money laundering compliance work, added a ‘couple hundred million plus’ to third-quarter costs, but he expects remediation costs to decline over time.

Anthropic catalogs Claude’s rule-bending on real websites

Anthropic has disclosed cases where Claude models took unintended actions on live websites during evaluations and internal use. In one case, a model exploited an injection flaw to run commands on a university server. Others involved submitting an invented tip through a Philadelphia Police Department form, and using exposed access tokens to reach fee-gated data. The company describes the impact as minimal, but it has now cut live internet access for all internal evaluations until it confirms its new detection tooling (which blocked every documented case in testing) works reliably.

West African scam centers targeted by tech giants and law enforcement

Meta, Coinbase, Google, Microsoft, OpenAI and ten other tech firms joined law enforcement from ten countries in a UK NCA-hosted operation against scam networks in Nigeria, Ghana and elsewhere in West Africa. Meta removed more than 3.7 million Facebook and Instagram accounts, pages and groups. Authorities arrested 28 people and rescued 117 trafficking victims who had been held in a scam compound, and the FBI said at least $11 million in potentially freezable funds was identified.

Burger King Japan app breach exposes 11 million member records

BK Japan, which runs Burger King’s official app in Japan, says hackers accessed the system holding the app’s member data between October 2 and October 7, exposing roughly 11 million member records. Exposed details include names, email addresses, phone numbers, dates of birth, gender and prefecture. The company says it doesn’t store card numbers and has so far found no sign that login passwords were taken.

AI-aided Chrome exploit for Windows takeover earns $250,000

XBOW has detailed a Chrome full-chain exploit, built by its researchers with help from OpenAI’s GPT-5.6 Sol, that let a malicious website take over a Windows machine. The final chain combines three bugs, all now fixed by Google. It was assembled in six days in August, after Google said it already knew about one of the flaws in the team’s first attempt. Google awarded XBOW a $250,000 bounty, while the compute required to develop the exploit cost $40,000.

KGB RAT pairs data theft with jump scares

Group-IB has analyzed KGB RAT, a Windows trojan sold on underground forums as a “fully undetectable” remote administration tool, with subscriptions ranging from $150 per month to $800 per year. Built on NjRAT foundations, it offers credential theft, keylogging, hidden VNC, ransomware and a crypto clipper. It also includes harassment features such as screamers, jump scares and forced blue screens. The developers push daily crypter updates to evade Windows Defender, and the samples Group-IB examined point to multiple operators using a shared builder.

Langflow flaw under attack weeks before CVE went public

Previdian says its honeypot sensors have recorded exploitation attempts targeting CVE-2026-105740, a critical Langflow vulnerability. The flaw lets any authenticated user run code on the server by adding an MCP server set to use the Stdio transport, because the supplied command is passed to bash without validation or sandboxing. The bug is fixed in version 1.9.0. Previdian first saw attempts on September 19, more than two weeks before the CVE was published on October 5, and has since logged over 1,800 attempts from 28 IP addresses in 16 countries.

Related: In Other News: $15K iCloud Spoofing Bugs, AI Policy Experts Phished, Adblocker Spies on AI Chats

Related: In Other News: AI Used in Korean Bank Breaches, Poem-Guided Botnet, Empire Admin Gets 40 Years