Aikido Security has disclosed that codexui-android, an npm package advertised as a remote web UI for OpenAI Codex with over 29,000 weekly downloads, has been silently exfiltrating users' Codex authentication tokens for the past month. Unlike a typosquat, the malware was embedded into a functional, actively-developed package roughly a month after publication to build trust; the GitHub repo stayed clean. The code reads ~/.codex/auth.json and ships the access_token, refresh_token, id_token, and account ID to sentry.anyclaw[.]store, a server masquerading as Sentry. The non-expiring refresh_token lets an attacker silently impersonate the developer indefinitely with full Codex account access. The package remains available; the npm account is 'friuns.'
VulnCheck says attackers are chaining three critical bugs (CVE-2026-28515, CVE-2026-28517, CVE-2026-28516) in openDCIM, an open-source data center management web app, to drop PHP web shells on exposed installs. All three rate CVSS 9.3 and cover missing authorization, OS command injection, and SQL injection. They can be combined over five HTTP requests to land a reverse shell. The activity comes from a single Chinese IP using what VulnCheck describes as a customized version of Vulnhuntr, a public AI-driven vulnerability discovery tool. The campaign is one of the first publicly documented cases of an open-source AI vuln scanner being repurposed for real-world exploitation.
The Gentlemen, the second most prolific public ransomware operation of 2026 with over 320 listed victims, has had its own internal database leaked. Check Point Research and others obtained the data after a breach of the group's hosting provider 4VPS exposed their Rocket backend. The leak unmasks roughly 9 named operators centered on an administrator known as zeta88 (aka hastalamuerte), who built the RaaS panel in three days using DeepSeek and Qwen AI coding assistants, runs payouts, and joins encryption events personally. Internal chats also confirm chain-victimization: in April the group hit a UK software consultancy and then weaponized stolen client credentials to compromise one of the consultancy's customers in Turkey.
Researchers at Cyera disclosed a critical bug in Ollama, the open-source tool that runs large language models locally on laptops and servers. The flaw, called Bleeding Llama (CVE-2026-7482), lets anyone with network access send a malformed model file and read raw process memory back - which typically contains API keys, environment variables, system prompts, and other users' chat history. Ollama ships without authentication by default, so an estimated 300,000 instances are exposed on the internet. Ollama 0.17.1 fixes it. Separately, Striga disclosed two unpatched Ollama Windows desktop flaws (CVE-2026-42248 and CVE-2026-42249) that chain into persistent code execution at login.
TrustStrike Labs