What I Check Before Letting an AI Agent Run a Task
Before giving an AI agent a task, I would review its permissions, inputs, action limits and verification process. A convincing answer is not evidence that an external action…
Read articleBlog
Investigations, cloud security, careers, and building in public.
A learner can remember terminology without being ready to investigate a problem. I describe how I would evaluate cybersecurity questions for reasoning, clarity and useful feedback.
Read article
PHP2Shell exposes PHP web servers to critical remote code execution risks. I explain the affected runtimes, fixed PHP versions, and the checks administrators should complete after patching.
Read article
Detecting an email address is easier than detecting every clue to a person’s identity. I explain how to evaluate redaction tools and review what survives.
Read article
Security headers provide useful browser protections, but a high score cannot establish that an application is secure. I explain what to check behind the grade.
Read article
Email headers can help explain delivery and authentication, but some fields can be forged. I describe how I separate trusted receiving evidence from a sender’s claims.
Read article
Reading a token’s claims does not make them trustworthy. I explain the separate jobs of decoding, signature verification, claim validation and authorisation.
Read article
Documentation becomes useful when another engineer can make a change or recover a service with it. I explain how I would reduce reliance on knowledge held by one…
Read article
Repeated requests for tables and views should become a controlled delivery process. I describe how I would automate warehouse changes while keeping access and recovery reviewable.
Read article
An aggregate dashboard can hide personal records on screen while its underlying connections still have broad access. I explain how to review that boundary.
Read article