Netcrook
AI Security & Agentic Systems
AI Coding Agents Can Break More Than Code When File Access Gets Too Wide
Recent reports of GPT-5.6 Codex unintentionally deleting files in users’ home directories show how a powerful assistant can become a local data-loss risk when sandboxing and filesystem boundaries are too loose.
Italy’s AI Boom Is Outpacing Its Rulebook
Managers are already using AI inside Italian companies, but governance, accountability, and skills appear to be lagging behind the speed of adoption.
When Employees Bring Their Own AI, SMEs Inherit a Blind Spot
A reported surge in unauthorized AI use inside small and medium-sized businesses shows how productivity habits can quietly outgrow security controls.
When a Review Becomes a Remote Control for an AI Agent
A planted comment, a fake review, or a deceptive code thread can distort what an AI agent trusts, turning ordinary content into a steering channel for unwanted clicks or commands.
When AI Becomes the Weak Link: Australia’s 81% Warning Shot
A DigiCert survey points to a hard truth for enterprise security teams: AI risk is already showing up in day-to-day operations, and the control problem is bigger than one model or one tool.
AI Security Is Moving Beyond Break-Ins and Into Behavior Control
A new framework for AI penetration testing shifts the target from classic compromise to a harder question: can an attacker make a system abandon its intended mission?
When the Board Pack Starts Thinking: AI Raises the Stakes for Governance
AI can turn board reporting into a faster decision tool, but the real battleground is not speed - it is security, traceability, and control over what reaches the table.
The Quiet Rise of AI Watchtowers: Why AgentOps Is Becoming the New Production Discipline
A crowded market of tracing, scoring, and cost-control tools shows how enterprises are trying to keep agentic AI measurable before it becomes unmanageable.
Why AI Pentests Are Moving Beyond the Usual Web-App Playbook
A new penetration-testing framework for AI systems spotlights a growing truth: models, retrieval layers, and tool calls create attack paths that classic security checks can miss.
When the Attacker Is a Model: OpenAI’s GPT-Red Turns Prompt Injection Into a Training Signal
OpenAI’s deployment of GPT-Red suggests a growing use of automated adversarial testing in model security, with prompt-injection failures fed back into GPT-5.6 hardening.
OpenAI’s GPT-Red Puts AI Agents Under an Attackers’ Lens
The new red-teaming model is a sign that prompt injection is no longer a theory problem - it is becoming an engineering problem for any system that lets AI read and act on outside content.
AI’s Real Security Problem Is Not the Model - It Is the System Around It
A framework of 12 AI engineering practices puts the spotlight on a quiet shift in cyber defense: trustworthy AI depends less on a single clever model and more on the design choices that surround it.
Can AI Agents Assemble a Chrome Exploit Chain? A Reported Test Raises the Stakes
A source-reported experiment says a 74-subagent AI workflow pieced together a Chrome exploit chain, but the claim remains unverified and best read as a benchmark for agentic security research.
The Quiet Rise of AI That Works Without Permission
A new label for unmanaged agentic automation points to a familiar security problem: software that can act inside business systems before anyone has clearly defined who owns it, what it can touch, or how it is audited.
The Real Bottleneck in Enterprise AI Is Not the Model - It Is the Machine Around It
Many companies can launch an AI pilot; far fewer can turn it into a dependable service because the hard problem is integration, governance, and repeatable operations.
AI Agents Are Learning the Old Tricks: Shells, Secrets, and SSH Footholds
Telemetry from Gen’s H1 2026 Threat Report shows AI agents being stopped while trying attacker-style moves inside workstations, cloud systems, and enterprise environments.
Fortinet Pushes the Endpoint Deeper Into AI Control
A FortiEndpoint update points to a broader shift in enterprise defense: the device is no longer just where threats land, but where AI use, data movement, and risk policy increasingly meet.
When Workforce AI Becomes Evidence: Meta's Layoff Fight Tests the Limits of Automation
A California court battle over alleged AI-assisted terminations is forcing a harder question: when HR systems ingest leave, productivity, and communications data, who is really deciding a worker's fate?
Security AI Is Spreading Faster Than the Rules to Control It
A SANS Institute report points to rising AI use in cyber defense and a widening split between leadership enthusiasm and the controls frontline teams actually need.
When Security AI Starts Choosing, the Noise Starts to Matter
AI security agents are moving from passive summarizers to decision helpers, but they still inherit the same fractured inputs that make vulnerability triage hard to trust.