High-signal AI/security/automation notes.
Varonis discovers Bluekit, a phishing kit under active development that embeds an AI assistant (Llama, GPT-4.1, Claude Sonnet 4, Gemini) alongside voice cloning and automated domain registration.
Material Security research finds 45% of organizations do nothing to monitor OAuth grants at scale, as AI tool proliferation creates persistent, invisible access tokens that bypass MFA and survive employee departures.
80% of security leaders flag unmanaged OAuth grants as critical risk, yet 45% of orgs do nothing to monitor them — AI tool proliferation makes this an urgent AI-ops security gap.
Critical unpatched RCE in PAN-OS User-ID Authentication Portal (CVSS 9.3) is being actively exploited; over 5,800 VM-Series firewalls exposed online, many protecting AI inference and training infrastructure.
A new ProjectDiscovery report finds that 60% of security practitioners struggle to keep up with the volume of AI-generated code, with secrets leakage and supply-chain risks as top concerns.
Trend Micro analyzes QLNX, a Linux rootkit/backdoor targeting developer workstations with credential theft across npm, PyPI, GitHub, AWS, Docker, and Kubernetes — enabling AI/ML supply-chain attacks.
Chrome 148 defaults to silently downloading a 4 GB on-device AI model and exposes it to any webpage through the Prompt API, raising privacy and consent questions for millions of users.
Benchmarking shows vision-based computer-use agents consume 45x more tokens than structured API agents, with high variance and failure rates that demand step-by-step prompt engineering.
Google overhauls Android and Chrome vulnerability rewards, scaling back payouts for AI-easier bugs while offering up to $1.5M for the hardest exploit chains.
NVIDIA patched two vulnerabilities in NemoClaw: CVE-2026-24222 (CVSS 8.6) lets prompt-injected content exfiltrate host environment variables through a sandbox initialization flaw, and CVE-2026-24231 (CVSS 5.9) enables SSRF via crafted endpoint URLs.
CIS publishes three companion guides mapping CIS Controls v8.1 to LLMs, AI agents, and Model Context Protocol environments, giving security teams a familiar framework for governing agentic AI.
Oasis Security discovered three chained vulnerabilities in Claude.ai — invisible prompt injection via URL parameters, data exfiltration, and an open redirect — creating a complete attack pipeline against default sessions.
CVE-2026-26268 shows how AI coding agents in Cursor IDE can be tricked into executing arbitrary code via malicious Git hooks embedded in untrusted repositories.
OpenAI and Anthropic met with the White House national cyber director, Congress, and federal agencies to propose deploying frontier AI models for government cybersecurity operations.