Blog
Notes from the work.
Short, practical reads from our team on security, cloud, and AI. Each one covers what changed, why it matters, and what to do about it, for the founders, executives, and teams making technology decisions.
19 pieces
HTTPS worked. The attacker controlled DNS.
Attackers hijacked three country-code registries and obtained valid certificates. Domain portfolios need monitoring and stricter issuance controls.
Read the piece
Your login gateway is part of the perimeter.
Citrix's NetScaler zero-day shows why SAML does not remove the appliance from the trust path. Patch the gateway, verify its configuration, and test access.
Read
An always-on agent is an account, not a feature.
Microsoft's Autopilot keeps working with its own identity and memory. Treat persistent AI agents as non-human identities from day one.
Read
A CAPTCHA does not need your terminal.
ClickFix turns a fake verification prompt into a malware installer. Train the tell, then put controls behind the person who still follows it.
Read
A system prompt is not a firewall.
Anthropic's cyber evaluations reached real systems through a test environment left open to the internet. The prompt described a boundary nothing enforced.
Read
A model upgrade is a production change.
GPT-6 Astra is faster, more capable, and built to act. That does not make it a drop-in upgrade for the agents already touching your business.
Read
Your emergency patch was not the finish line.
PaperCut's first emergency fix was followed by a second after researchers found bypasses. Patch now, but do not confuse updated software with a clean server.
ReadStart with a conversation.
Our team will respond with a considered view of where to begin.