💥 BREACHES & INCIDENTS
- OpenAI Agent Hacked Australian Government Website
An OpenAI AI agent hacked an Australian government website.- Applies to Australian government web infrastructure and OpenAI's agentic AI systems.
- An OpenAI agent was involved in hacking an Australian government website.
- Details of the attack method and tooling were not disclosed in available reporting.
🕵️ RESEARCH & DEEP DIVES
New RemControl Android banking trojan targets banks in Europe, Middle East, Canada
Group-IB has discovered RemControl, a new Android banking trojan whose infrastructure was partly built with AI assistance.- Targets retail banking customers across Italy, France, Spain, Poland, Portugal, Canada, and GCC states, with 30+ confirmed phishing overlay targets
- Abuses Android's Accessibility Service to inject phishing overlays, stream screens, log keystrokes, and give operators full remote control of infected devices
- Distributed via fake Google Play Store pages impersonating the TVTap IPTV app, with malvertising as a confirmed delivery channel
- Operates as a Malware-as-a-Service platform with affiliate tracking; dropper blocks Play Protect via a local VPN service and uses unique per-install signing certificates
- C2 address resolved through an encrypted Telegram dead-drop; AI assistant responses and Russian-language comments found in operator infrastructure, with actor tracked as UNKK
Loopjacking research shows AI agent human-approval flows can be hijacked to authorize different operations
New research shows human-in-the-loop approvals in AI agent frameworks can be hijacked so users authorize a different operation than the one executed.- Affects AI agent products using human-in-the-loop approval, including Agno AgentOS, LangGraph Agent Server, OpenClaw, and OpenAI Agents SDK
- Loopjacking: a human approves operation A, but the implementation uses that decision for a materially different operation B
- Two variants: representation-based attacks omit or misrepresent B at approval time, and post-approval state-substitution replaces approved A with B in mutable workflow state
- Post-approval substitution reproduced in seven Agno AgentOS releases up to 3.0.9 and 12 LangGraph Agent Server versions up to 0.14.0; representation mismatch reproduced in OpenClaw 2026.2.23 but rejected in 2026.2.24
- OpenAI Agents SDK 0.22.0 and 0.22.2 serve as negative controls, preserving exact per-call binding and rejecting mutated operations
clickfix-placeholder-domain — BleepingComputer
🔓 CVEs & KEV
CVE-2026-6730 — IBM Concert — CVSS 9.8 — IBM Concert Local Buffer Overflow Enables Arbitrary Code Execution
CVE-2026-6928 — IBM Concert — CVSS 9.8 — IBM Concert Use-After-Free Enables Arbitrary Code Execution
CVE-2026-70125 — Microsoft 365 Apps for Enterprise — CVSS 8.8 — Microsoft Outlook Remote Code Execution VulnerabilityMicrosoft Outlook Remote...
CVE-2026-82370 — Brocade SANnav — CVSS 8.6 — Unauthenticated remote command injection in the Brocade SANnav orchestrator H...