Threat actors

Grey market relay service disables OpenAI Codex and Claude Code sandboxes and reroutes enterprise data

Symantec's Threat Hunter Team identified a relay service active since at least early 2026 that has touched about a dozen organisations, mostly in mainland China. Users install the genuine Codex or Claude Code tool, then run a fileless setup script from a lookalike domain that rewrites the config to point at relay servers, stores the relay key as a user-level variable, sets sandbox mode to danger-full-access, enables network access and turns off telemetry. This is not a flaw in either product, but the relay can see every prompt, file and secret and could alter replies to run commands with the user's rights. No such abuse was observed.

Chinese state-linked group Vapor Panda uses AI-powered ARTEX framework against high-value targets

CrowdStrike attributes use of ARTEX, a Go-based AI-powered penetration testing framework, to a China-based targeted intrusion adversary it tracks as Vapor Panda, previously VexatiousTemper. The headline refers to more than 200 organisations worldwide, but the supplied text is truncated and gives no detail on victims, sectors or techniques. The activity is described as aimed at high-value targets simultaneously.

Apple warns users in 99 countries of mercenary spyware attacks, over 150 countries notified to date

Apple issued a new round of threat notifications to targeted users in 99 countries, bringing its total to over 150 countries. Apple does not attribute the attacks to any specific attackers or regions. Citizen Lab and Access Now advise recipients to enable Lockdown Mode and contact Access Now, and warn that the absence of a notification does not rule out infection. The previous major batch was sent in August, and the latest one comes shortly before final negotiations on industry standards under the voluntary Pall Mall Process.

Teams help desk impersonation leads to AD CS ESC1 and ESC8 certificate abuse in two intrusions

Microsoft Threat Intelligence investigated two human-operated intrusions that began with Teams messages and calls impersonating help desk staff, leading users to approve Quick Assist sessions. The attackers staged portable Python, Bun or Java runtimes, performed discovery and probed AD CS. In Case 1 an ESC1-vulnerable template gave certificates for privileged accounts and DCSync. In Case 2 PetitPotam and an ESC8 relay gave a domain controller certificate.

Palo Alto Networks firewalls compromised in chained attacks using PHP2Shell

Cybersecurity Dive reports that Palo Alto Networks firewalls were compromised in chained attacks that used PHP2Shell. The AI-enabled campaign hit two dozen targets, including sites linked to the U.S. government. The article excerpt gives no further technical detail.

Unauthorised Codex and Claude Code relay service disables sandboxing at 12 organisations

Symantec's Threat Hunter Team identified an unauthorised relay service that installs the genuine OpenAI Codex and Anthropic Claude Code command-line tools and rewrites their configuration to disable sandboxing. It has affected 12 organisations since early 2026, mostly in mainland China. Because these AI coding agents can read files, write code and run commands, removing sandboxing widens what they can do on a host.

Cleared defence contractor pleads guilty to aiding North Korean remote worker scheme

John Allen Kurrle Jr., 62, pleaded guilty to conspiracy to commit wire fraud for letting co-conspirators use his identity and clearance to apply for 88 US jobs, leading to nine offers between June and October 2025. He hosted laptops from victim companies at his Virginia home so operatives could access them remotely from abroad. A Californian firm working on a generative AI project for a US military component caught the mismatch on a video call and fired the worker before any contract work began.

DeceptiveDevelopment hackers steal developer credentials to poison GitHub repositories and software packages

ESET reports that DeceptiveDevelopment, a North Korea-aligned group, used stolen Git credentials, SSH keys and tokens between April and September 2026 to alter public and private repositories, often rewriting commit history. Poisoned code appeared on npm, the Go module mirror, Packagist, PyPI and browser extension marketplaces, and the group deployed the OmniStealer infostealer and the DEVPOPPER remote access trojan.

Coruna and DarkSword iPhone exploit kits spread via 608 web domains and a backdoored npm package

Zimperium zLabs documented 608 domains delivering Coruna-family code, which selects an exploit chain by iOS version: Coruna for iOS 17.2.1 or older and a DarkSword-based chain for iOS 18. Delivery is through decoy lure sites with hidden frames, a backdoored art-template npm package reported by SafeDep, and a fake jquery.min.js loaded from a look-alike Google domain. The malware steals data such as Telegram storage and the Apple Notes database, and deletes crash logs and Safari data to hide its tracks. Every CVE in the chains is already patched.

Unknown APT uses NetamLoader to attack a Win10 OS customized for governments in Asia

ESET APT Activity Report Q2 2026–Q3 2026

Talos IR Q3 2026: Help desk vishing grows, ransomware rises to 27 percent of engagements

Talos IR found phishing was the top initial access method at just over one-third of engagements, down from over 50 percent, with vishing and help desk impersonation used to get victims to allow remote access via legitimate RMM tools. Exploitation of public-facing applications, mainly edge devices, accounted for one-third. Ransomware and pre-ransomware activity reached 27 percent, involving The Gentlemen, Qilin, Rhysida, Interlock and the newer VYPR, with EDR disabled via vulnerable drivers or WDAC.

Leak site publishes data on more than 41,000 DHS employees obtained by hacktivists

Volt Typhoon used covert Silent Train relay network to exploit zero-days in management platforms, Lumen reports

Lumen's Black Lotus Labs reports that Volt Typhoon used an operational framework called Silent Train, first seen in November 2023, to chain compromised routers, devices and enterprise servers as covert relays. The framework supported four confirmed and one likely zero-day campaign against management platforms including N-central, Versa Director, Cisco Catalyst SD-WAN, VeloCloud Orchestrator and Quest KACE. Victims included telecommunications, satellite communications, defence-linked universities and essential service operators, with Hawaii, California and Texas repeatedly targeted. Lumen assesses the activity as positioning for possible Indo-Pacific crises rather than evidence of disruption.

Tracking AI adoption in North Korean threat actor tradecraft

Hacktivist leak exposes records of more than 41,000 DHS staff, including undercover ICE agents

A group calling itself the Department of Peace leaked records on more than 41,000 DHS employees and contractors, covering DHS and seven component agencies including ICE and CBP. The data includes email addresses, phone numbers, office addresses, references to internal systems and information on undercover and surveillance programmes. Distributed Denial of Secrets published the archive, named Icepick, and says it comes from DHS's Okta login system.