Briefing — 3 September 2026

🎧 Subscribe to the podcast feed

Security News

METR Report on OpenAI / Hugging Face Hacking Incident (Hacker News / METR) — Independent researchers document how an attacker breached AI lab infrastructure at both OpenAI and Hugging Face, highlighting AI development pipelines as high-value targets.

An AI-Assisted Cyber Attack: Inside a Unit 42 Investigation (Unit 42 / Palo Alto Networks) — Real incident response case: attacker used frontier AI and agentic frameworks to compress two weeks of red-team tradecraft into under 10 hours, using 50+ MITRE ATT&CK techniques including hijacking the victim's own AI infrastructure.

Claude Mythos only model to complete full cyber kill chain, experts say (The Register) — Researchers document the first AI model to autonomously complete a full cyber kill chain in testing, marking a milestone in offensive AI capability.

SonicWall SMA 1000 Zero-Days Enable Unauthenticated RCE (Dark Reading) — Critical zero-day vulnerabilities in widely-deployed SonicWall SMA 1000 remote access appliances allow unauthenticated remote code execution — treat as urgent.

Hackers exploit Sangoma Switchvox flaw to deploy reverse shells (BleepingComputer) — Active exploitation of a vulnerability in Sangoma's Switchvox VoIP platform to deploy reverse shells on affected systems.

WordPress backup plugin flaw exposes millions of sites to takeover attacks (BleepingComputer) — Critical vulnerability in a widely-used WordPress backup plugin enables full site takeover across millions of installations.

Impersonating IT support: how threat actors turn a remote session into enterprise-wide access (Microsoft Security Blog) — Threat actors impersonate IT support staff to gain remote sessions, then pivot to enterprise-wide compromise.

Hunting MacSync Stealer infrastructure through behavioral pivots (Microsoft Security Blog) — Threat hunting pivots to identify MacSync Stealer infrastructure — a macOS infostealer targeting enterprise credentials.

I rented a car, and within hours, my driver's license was for sale (Ars Technica) — 153 million driver's licenses available on a dark web marketplace via routine data broker activity, no breach required.

AI's Vulnerability Surge May Be More Manageable Than First Feared (Dark Reading) — Analysis suggests the rapid growth in AI-related vulnerabilities may be addressable with existing security frameworks.

OpenLeash Adds a Human Check to Risky AI Agent Actions (SecurityWeek) — New tool inserts human approval gates for high-risk agentic AI actions, addressing a key concern in AI security.

UK & Critical National Infrastructure

⚠️ Two UK organisations appear in this cycle's ransomware victim list.

🇬🇧 Manchester Airports Group (MAG) — posted by Fulcrumsec. MAG operates Manchester, London Stansted, and East Midlands airports. Significant transportation CNI target.

🇬🇧 Marlborough Partners — private equity firm, posted by Anubis. Financial sector hit.

No new NCSC advisories in the past 48 hours.

Broader CNI context: Qilin hit Grayson Rural Electric Cooperative (US) and Direwolf targeted PTT Oil and Retail Business (Thailand) this cycle. UK CNI operators in energy and transport should assess exposure against the Krybit threat model, given the group's 48% infostealer-linked initial access rate.

Ransomware Victims (48h)

43 victims · 19 groups

Today's Picks

• An AI-Assisted Cyber Attack: Inside a Unit 42 Investigation — Documented real-world incident of AI-accelerated intrusion; essential reading for defenders.
• From GPO to Microsoft Intune: A practical guide to cloud-first policy management — Solid methodology for migrating away from Group Policy to Intune.
• Obscura — lightweight headless browser for AI agents — ~30MB alternative to Chrome for web automation in agentic workflows.