LodeHQ
AI news for people who shipFree, by email

Infosec · Wednesday, 1 July 2026 · Issue 32 · 9 stories

Attackers probe Kemp RCE, Chrome sandbox falls to one bug

Vulnerabilities & Exploits

Fake Sentry Error Lets Attackers Hijack AI Coding Agents

Tenet Threat Labs proved that a single forged Sentry bug report can make AI coding assistants, Claude Code, Cursor, Codex, execute attacker‑controlled code, spilling AWS keys, GitHub tokens and other secrets. The exploit works across hundreds of firms, including a Fortune 100, because agents trust telemetry as native system output, bypassing EDR and firewalls. Developers must treat external logs as untrusted and apply the newly released hardening configs.

tenetsecurity.ai ↗17 min read at source

Threats & Malware

ToddyCat’s Umbrij steals Gmail OAuth tokens for silent, persistent access

Kaspersky uncovered Umbrij, a custom tool the ToddyCat APT uses to hijack Gmail accounts via OAuth token theft. By attaching to a Chromium browser’s remote‑debug port, the malware silently grabs a “shadow token” and gains persistent API access to email and other Google services. Detect it via unusual scheduled tasks, DLL sideloading, and Kaspersky’s heuristic verdicts.

securelist.com ↗17 min read at source

Browser-Only Ransomware Proves AI‑Generated Code Can Encrypt Files Without Installing Malware

Check Point Research turned an LLM hallucination about a nonexistent 'browser-only ransomware' into a working proof‑of‑concept that encrypts files wholly inside the browser using JavaScript and the File System Access API. The attack requires no downloaded executable, sidestepping traditional endpoint defenses and expanding the threat surface for browsers that grant file‑system permissions.

Breaches & Industry News

Privacy, Policy & Governance

Research & Tools

You’re caught up.

That was all 9 stories in the 1 July issue.