Free weekly newsletter
AI Security Intelligence
What actually breaks, and what to do about it.
Contributor, OWASP Top 10 for Agentic Applications 2026 · CISSP · PhD · 15+ years enterprise security
One issue every Wednesday. Practitioner-grounded coverage of LLM and agentic security: prompt injection patterns, MCP attack chains, RAG security, OWASP Top 10 for Agentic Applications 2026. Written for engineers and security teams who want to know what's actually breaking, not vendor positioning.
What you'll get
- One issue per week, every Wednesday. Roughly 600 to 800 words. Reading time: 4 to 6 minutes.
- Practitioner-led format. Each issue anchors on something I'm actively building or testing myself. News stories appear when they're evidence for a practitioner question, not as the main event.
- No vendor PR. I don't write about vendors who haven't published something substantive. I don't run sponsored issues.
- No em dashes. (You'd be surprised how often this is asked.)
Recent issues
[AI Sec Intel] #31 The VM held. The agents left through DNS
Oct 11, 2026
Anthropic, Google and OpenAI all disclosed agents reaching systems they were walled off from.
Read on Beehiiv →[AI Sec Intel] #30 A human approved a transfer of 20. The framework sent 2,000
Oct 4, 2026
Human approval only counts if the thing you approved is the thing that runs. In two popular agent frameworks it isn't.
Read on Beehiiv →[AI Sec Intel] #29 MCP deleted the session, and the session was doing your access checks
Sep 26, 2026
If a tool takes a cart ID or a workflow ID, anyone holding that ID can use it unless your code checks who is calling.
Read on Beehiiv →[AI Sec Intel] #28 A quarter of the talks at MCPCon Europe were about security. Not one said the fix was a better prompt
Sep 19, 2026
Zalando built an identity broker for its agents, open sourced it, and runs it in production, so agent identity is now a solved and shipping problem.
Read on Beehiiv →Who reads this
- Engineering leads at companies deploying LLM products
- Security engineers and architects working on AI features
- Privacy and compliance professionals tracking the AI Act intersection
- Researchers and consultants in the AI security space
- A growing number of subscribers from US enterprise AI vendors as the EU AI Act enforcement approaches
About the author
I'm Amine. I help teams build LLM systems that don't become security incidents. My background is in securing critical systems across regulated industries. More recently I've been deep in how enterprise teams build and deploy LLM-based systems, and where those systems break. That combination is where this newsletter comes from.
More about me →Working together
When an issue hits close to home
The newsletter is where I write about what's breaking. Molntek is where I help fix it. If one of these issues describes something you're shipping right now — an agent, an MCP integration, a RAG pipeline you're not sure about — that's exactly the work I do at Molntek. No pressure, and subscribing never puts you on a sales list.
Common questions
Is it really free?
Yes. The newsletter is unsponsored and won't move behind a paywall.
How do you handle my email?
Email used to send the newsletter only. Hosted on Beehiiv. No third-party sales. One-click unsubscribe in every issue.
Will I get sales emails?
No. The newsletter list and Molntek client list are separate. If you want to talk to me about consulting work, you contact me. I don't reach into the newsletter list to sell.
What if I want to read past issues?
Public archive at newsletter.aminrj.com. No subscriber gate.
Can I share an issue?
Yes. Forward, post, quote with attribution. The point is reach.