The CyberJuice blog
The latest insights on compliance, security and AI, from people doing the work.
Articles
Permission reviews aren’t a judgment call. They’re list-matching, and that’s the part worth automating
Permission reviews feel like judgment calls but are mostly manual list-matching, which is why they’re a good candidate for automation.
Your CPR Number Has 10,000 Possible Endings, and That Was the Whole Plan
A four-digit suffix was never a real secret, and treating it as proof of identity was the design flaw all along.
A Danish accent used to mean something. AI voice cloning just erased that assumption
A Danish accent no longer proves a caller is who they claim to be, and banks still treat partial, guessable details as sufficient identity checks.
The AI sidebar in your app only knows what that app knows
An in-app AI sidebar can only answer from that app’s own data, so cross-system questions send it guessing while connector-based assistants can actually see across everything you’ve connected.
The MCP connector list tells you a tool shows up, not whether it does anything useful
Having a connector listed doesn’t mean the feature you actually want is reachable through it, Zapier’s Zaps being the clearest case.
Block WeTransfer and People Just Use Their Hotspot Instead
Blocking a tool just pushes the workaround out of sight. Watching real behavior, or asking about it directly, finds the actual problem instead.
Your security policy describes a company that doesn’t exist
Templates describe an idealized company that enforces everything perfectly. Yours should describe what you actually do, with the gaps written down separately.
The AI Agent Didn’t Destroy Your Database. Your Access Controls Did.
An AI agent dropping tables is not an AI problem, it is a sign nobody scoped its database permissions properly.
The Policy Says No Unapproved Cloud Sharing. Nobody Checked If the Firewall Agrees.
A written rule against unapproved cloud sharing means little if the firewall was never configured to detect or block it.
Clean Code, Broken Architecture: Why Review Can’t Catch a Bad Design Decision
Clean, review-passing code can still sit on a broken design, because review checks execution, not whether the request should have been made at all.
We Moved Our Content Out of the CMS So AI Agents Could Actually Read It
Content, structure, and links moved out of the CMS into Markdown in the codebase so AI agents can read and edit them with proper version control.
Meet Cyberjuice × Claude: Your cybersecurity compliance just got conversational 🚀
What if managing ISO 27001 was as easy as asking a question? No more searching through endless documentation.
See it with your own company in mind
See CyberJuice working on the things your customers and auditors ask for.
Book a demo