THE LUCIN BLOG

Field notes on agent security.

Teardowns of real incidents, the graph theory behind the detectors, and the benchmark numbers with the commands that produce them.

POSITION
06 AUG 2026
9 min read
Every agent security tool grades its own homework

Every agent security tool grades its own homework. A proposal for an open benchmark, and our own scores on it

Read the write-up →
METHOD
05 AUG 2026
11 min read
An information-flow model for AI agents, and what it provably cannot decide

The lattice, the reachability query, and the min vertex cut, in 985 lines. Plus the one question the static model provably cannot answer, which is where the runtime half has to begin.

Read the explainer →
TEARDOWN
04 AUG 2026
8 min read
Your agent framework has three CVEs. Here is what each one actually does.

Three LangChain and LangGraph CVEs, what each one actually does, and why the kill chain you have read about does not exist

Read the teardown →
ESSAY
03 AUG 2026
7 min read
Prose is executable

The payload in CVE-2026-25724 was a comment in a README, which makes natural language an execution path

Read the write-up →
PROOF
01 AUG 2026
9 min read
We publish our misses. Here are all twelve.

76% recall, the twelve cases that get past us, and the detector we switched off on purpose

Read the write-up →
PROOF
31 JUL 2026
7 min read
I published my false-positive rate and the command that regenerates it. Here is why almost nobody does.

Every scanner publishes what it catches. Almost none publish a command you can run to check. Here is my false-positive count, the number that argues against it, and both commands.

Read the write-up →
TEARDOWN
29 JUL 2026
9 min read
17,600 actions, 2.5 days, nobody watching: anatomy of the Hugging Face agent breach

A malicious dataset, a tool that runs code, a credential read, an outbound call. Four ordinary tools, none of them a mistake on its own. Drawn as a graph, they are one edge, walked 17,600 times.

Read the teardown →
METHOD
27 JUL 2026
7 min read
The lethal trifecta for AI agents, and how to actually cut it

Private data, untrusted content, external reach. Any one is fine, all three is an incident. Open your tool file: this is the ten-minute pass that tells you which edges to cut and which to gate.

Read the explainer →
TEARDOWN
25 JUL 2026
9 min read
The attack chain started by compromising a security scanner

A credential harvester in a library with 95 million monthly downloads. The way in was a security scanner, which is the reason this post spends a section on why you should not trust mine.

Read the teardown →

More posts as we publish them. Every claim links to the command that reproduces it.

Check your own agent for this shape.
Free, MIT, 30 seconds.