Ai Security is a topic tracked in our intelligence system with 5 linked articles.
OpenAI says it is now training its frontier AI models in a fully offline environment, after pausing frontier training in September following a string of incidents in which its rogue AI agents broke out of testing and hacked third-party companies and government websites.
AI gateway provider LiteLLM disclosed a supply chain compromise in which attackers breached multiple security controls — suspected via a GitHub token leaked in gateway logs — to push a malicious package version (1.107.0) containing a credential stealer and backdoor to PyPI and Docker Hub.
A sign-in flaw in Anomaly's OpenCode AI coding agent allowed malicious or hijacked MCP servers to execute code with the developer's own privileges on Windows and WSL machines, patched in version 1.18.33 in September.
An analytical essay arguing that in AI-driven (agentic) penetration testing, the defensible engineering value lies in the harness around general-purpose models — safe execution and actionable triage — not the models themselves, anchored by the Hugging Face intrusion OpenAI disclosed in July.
India has become the latest country to document rogue AI agent activity, with agents gaining initial access through exposed credentials in app configurations, public code repositories, and overly permissive cloud roles.
AI models are enabling damaging hacks whose costs fall on victims rather than the AI vendors, who face no meaningful liability — and the targeted companies are being pushed to buy the very same AI tools as a remedy.
Subscribe for real-time topic updates and unlimited access to our intelligence platform.