zybeBlogNewsletterJoin the launch list

Blog

What we see on the endpoints

The Meta AI agent data leak: what happened, and what your security team should change
The Meta AI agent data leak: what happened, and what your security team should change

A Meta AI agent posted bad advice without approval, and data was exposed to staff for two hours. What was reported, and four controls to check at your company.

30 October 2026

 Amazon's Kiro outage: was an AI coding bot to blame, and what should your team change?
Amazon's Kiro outage: was an AI coding bot to blame, and what should your team change?

What the FT reported, what Amazon says, and the access control lesson for any team using AI coding agents, with five things to check this week.

30 October 2026

 Claude Code --dangerously-skip-permissions: what it means for your security team
Claude Code --dangerously-skip-permissions: what it means for your security team

One flag turns off Claude Code approval prompts on a developer laptop. What it exposes, why you can't see it today, and the policy to set for your company.

30 October 2026

 OWASP Top 10 for Agentic Applications: what each risk means if your company runs AI agents
OWASP Top 10 for Agentic Applications: what each risk means if your company runs AI agents

The ten OWASP agentic risks in plain language, which ones apply when your staff run AI agents, and the evidence to collect for each one.

30 October 2026

 Claude Code hooks: the code that runs on every tool call, and who put it there
Claude Code hooks: the code that runs on every tool call, and who put it there

Claude Code hooks run on every session and tool call with the developer's permissions. Where they come from, how one persists, and how to see every hook on every machine.

7 October 2026

 AI acceptable use policy: the section yours is missing for AI coding agents
AI acceptable use policy: the section yours is missing for AI coding agents

Most AI acceptable use policies cover chatbots. Here are the seven clauses to add for AI coding agents, a section to copy, and how to know it is followed.

3 October 2026

 Codex CLI security: what it is, what it can reach, and how a company governs it
Codex CLI security: what it is, what it can reach, and how a company governs it

Codex CLI for the security team: sandbox and approval modes, where its config lives, what it reads on a laptop, and the file a developer cannot override.

29 September 2026

 Claude Code permissions: the model, the three switches, and the standard to set
Claude Code permissions: the model, the three switches, and the standard to set

Claude Code permissions for the security team: allow, ask and deny rules, the six modes, the five settings files, and a managed baseline to deploy.

29 September 2026

 ISO 42001 for AI agents: what the Annex A controls ask and how to prove it
ISO 42001 for AI agents: what the Annex A controls ask and how to prove it

What ISO/IEC 42001 asks of a company whose developers run AI agents, control by control from Annex A, and the endpoint evidence that proves it.

27 September 2026

 Cursor rules: the instructions your agents follow that nobody reviews
Cursor rules: the instructions your agents follow that nobody reviews

Cursor rules are text the agent follows, committed by anyone. How a poisoned rule reaches every developer, and how to review and lock rules down.

27 September 2026

 Claude Code security: what it can reach, what to lock down, what to watch
Claude Code security: what it can reach, what to lock down, what to watch

What Claude Code can reach on a developer machine, how its permissions get switched off, and the five managed settings a security team should lock down first.

26 September 2026

 Cursor vs Claude Code: what each can reach on a developer machine, and what to lock down
Cursor vs Claude Code: what each can reach on a developer machine, and what to lock down

Cursor vs Claude Code from the security side: where each keeps its config, how each stops asking, how MCP and hooks load, and what to lock down.

26 September 2026

 AI agent security: what it is, where it happens, and the four controls
AI agent security: what it is, where it happens, and the four controls

AI agent security when the agents are Claude Code, Cursor and Codex on developer machines: what goes wrong, the real incidents, and the four controls.

26 September 2026

 The OWASP LLM Top 10 on a developer laptop: what each risk looks like and how to see it
The OWASP LLM Top 10 on a developer laptop: what each risk looks like and how to see it

The OWASP LLM Top 10 applied to Claude Code, Cursor and Codex on developer machines: what each risk looks like and the evidence for it.

26 September 2026

 EU AI Act compliance for AI coding agents: what it requires and how to prove it
EU AI Act compliance for AI coding agents: what it requires and how to prove it

What the EU AI Act asks of a company whose developers run AI coding agents, article by article, and the endpoint evidence that proves it.

25 September 2026

 Shadow AI agents: how to find every AI coding agent your developers run, today
Shadow AI agents: how to find every AI coding agent your developers run, today

Shadow AI agents run on developer machines with full access and no inventory. How a security team finds every one, and what to do first.

23 September 2026

 AI agent governance: how a security team sets the rules for AI coding agents across the company
AI agent governance: how a security team sets the rules for AI coding agents across the company

AI agent governance for the CISO: the four decisions to make about AI coding agents, how to enforce them on every endpoint, and what the auditor gets.

22 September 2026

 MCP security: what an MCP server can do on a developer laptop, and how to review one
MCP security: what an MCP server can do on a developer laptop, and how to review one

MCP server security explained: what an MCP server can reach on a developer laptop, three real 2025 incidents, and how to review one before it runs.

21 September 2026

 Best AI Agent Security Tools for Dev Teams in 2026
Best AI Agent Security Tools for Dev Teams in 2026

Comparing the top ai agent security tools for developer machines in 2026: endpoint coverage, prompt injection blocking, pricing, and what each platform misses.

20 September 2026

 AI Agent Security Risks Every Dev Team Must Know
AI Agent Security Risks Every Dev Team Must Know

AI agent security is now a frontline problem for dev teams. Here are the biggest risks your agents create on developer machines and how to stop them.

19 September 2026

 AI agent security: the agentic endpoint is your new boundary
AI agent security: the agentic endpoint is your new boundary

AI agent security explained: what AI coding agents can read, run and reach on a developer laptop, why EDR misses it, and how to secure the agentic endpoint.

10 September 2026

 Prompt injection attacks on AI agents: a real example, and how to stop them
Prompt injection attacks on AI agents: a real example, and how to stop them

A step-by-step prompt injection attack on an AI coding agent: the exact tool calls, why it works, and how to stop it on the endpoint.

3 September 2026

 What credentials can an AI coding agent access? A real count
What credentials can an AI coding agent access? A real count

AI agent credential access, counted on one real laptop: forty-six secrets an AI coding agent could read, why that is normal, and how to control it.

27 August 2026