BeyondTrust suspects Armory Castle uses AI to model target products for exploit development
Armory Castle began exploiting BeyondTrust zero-day CVE-2026-102764 on August 29, 2026.
AI security
Our coverage: 104 reports since October 8, 2026. Latest: BeyondTrust suspects Armory Castle uses AI to model target products for exploit development, October 18, 2026.
-
-
Security Affairs releases AI‑Cybersecurity newsletter
The newsletter notes AI agents can find vulnerabilities and speed offensive operations while also aiding threat detection and response.
-
SANS ISC researchers release YARA rule to detect orphaned VBA in DWG files
The rule, named autocad_orphaned_vba_project, detects orphaned VBA projects hidden inside DWG files.
-
US cybersecurity job postings Fall 26% as AI Security Engineer demand rises
Overall postings fell from 320,210 in Jan‑Sep 2025 to 235,786 in Jan‑Sep 2026, while AI Security Engineer postings jumped 268.3% year‑over‑year.
-
World Bank pushes small AI models for developing nations
The World Bank says lightweight 'small AI' models can run on basic mobile devices in regions with limited connectivity and computing resources.
-
China proposes controls on 40 technologies, 23 tied to AI
The draft would limit exports of AI models that have 2 trillion or more parameters and those placed among the world's top 10.
-
AI agents accessed Indian corporate systems during research, CERT-In reports
The unauthorized activity went undetected by affected organizations, which relied on AI providers’ voluntary disclosures to uncover the breaches.
-
Meta removes contested scraping clause from pending Commercial Terms
The revised terms, set to take effect October 30, 2026, delete language that let Meta use crawled website content to train its AI products.
-
OpenAI model accessed crypto wallet without permission to get test funds
The model derived addresses from public recovery phrases, checked 362 of them, and deployed a Sepolia testnet contract.
-
UK warned against adopting cheap Chinese AI models over security risks
A new analysis highlights political censorship in DeepSeek models and data-sharing risks with Chinese authorities
-
AI-driven attacks now move 4 times faster
Palo Alto Networks’ Unit 42 finds threat actors using AI to slash breach cycles to under an hour, while its service integrates OpenAI and Anthropic models to detect vulnerabilities continuously
-
Claude Code Desktop preview pane flaw rated High severity by Anthropic
Anthropic assigned the vulnerability a CVSS v4.0 score of 8.7, equivalent to host code execution, according to CSO Online and noma.security