The 30 AI products Securaity tracks for security advisories

Securaity follows the vulnerability records of 30 AI software products every day. This post lists all 30 with their vendors and the kind of tool each one is. It also shows how many advisories each product had in the 180 days to 2026-10-27, so you can see which of the tools your organization runs have been busy. The figures come from our product list and database, read on 2026-10-27.

What the list covers

Most of the 30 products are open-source or self-hosted tools. That means your organization, your developers or your IT provider installs and runs them on your own machines or servers. They fall into 16 kinds, including:

• Frameworks and libraries for building models and AI apps
• Inference servers that run models
• Chat interfaces
• LLM app builders
• Workflow automation
• Vector databases
• Notebooks

The largest kinds are inference servers and vector databases, with 4 products each.

Hosted AI services are not on this list. These are the chat assistants and other tools you use in a browser. We cover them in the tool safety profiles under /tools/.

Each tracked product has its own page under /vulnerabilities/. The page lists its advisories, newest first, and has an RSS feed.

How we match records to products

We look for each product in 2 public databases:

• National Vulnerability Database (NVD): we match each product by its CPE identifier, which is a standard name for a piece of software. Two products have no CPE entry. For those, we match an exact product phrase near the start of the record’s description.
• GitHub Advisory Database: we match by package name in the pip, npm and Go ecosystems.

A product page stays out of search engines until it has at least 5 advisories. You can still open any product page from the list at any time. Our methodology page has more detail on matching.

The full list, with advisory counts

The counts below are published advisories from records that their source first published in the 180 days to 2026-10-27.

Product	Vendor	Kind	Advisories
llama.cpp	ggml	Inference server	13
NVIDIA Triton Inference Server	NVIDIA	Inference server	17
Ollama	Ollama	Inference server	3
vLLM	vLLM project	Inference server	67
Chroma	Chroma	Vector database	5
Milvus	Milvus	Vector database	1
Qdrant	Qdrant	Vector database	0
Weaviate	Weaviate	Vector database	2
AnythingLLM	Mintplex Labs	Chat interface	4
Open WebUI	Open WebUI	Chat interface	137
text-generation-webui	oobabooga	Chat interface	0
Dify	LangGenius	LLM app builder	2
Flowise	FlowiseAI	LLM app builder	88
Langflow	Langflow	LLM app builder	154
Keras	Keras	Machine learning framework	9
PyTorch	Linux Foundation	Machine learning framework	0
TensorFlow	Google	Machine learning framework	0
LangChain	LangChain	LLM application framework	3
LlamaIndex	LlamaIndex	LLM application framework	0
Gradio	Gradio	Machine learning app framework	3
Streamlit	Snowflake	Machine learning app framework	1
LM Studio	Element Labs	Desktop LLM app	0
Ray	Anyscale	Distributed computing framework	2
ComfyUI	Comfy	Image generation tool	0
ONNX Runtime	Microsoft	Inference runtime	0
LiteLLM	BerriAI	LLM gateway	27
MLflow	LF Projects	MLOps platform	17
Hugging Face Transformers	Hugging Face	Machine learning library	4
Jupyter	Project Jupyter	Notebook environment	18
n8n	n8n	Workflow automation platform	150

The numbers at a glance

• 30 products across 16 kinds
• 727 advisories in all during the window
• 22 products have at least 1 advisory
• 8 products have none in the window
• 12 products have 5 or more

Langflow (154), n8n (150) and Open WebUI (137) have the highest counts. Flowise (88) and vLLM (67) come next.

What these numbers do and do not show

Read the counts as a record of public activity, not as a score.

This is not a ranking or a recommendation. A high count does not mean a product is a bad choice, and a low count does not mean it is a good one. Inclusion on the list means only that we follow the product’s records.

No advisories does not mean safe. Eight products had no advisories in this window. That tells you nothing was published about them in these 2 databases during these 180 days. It does not prove the software has no flaws.

The window is 180 days. Older advisories are outside these counts, even when they still matter for an unpatched install.

Only public records count. We work from the NVD and the GitHub Advisory Database. A problem that was fixed quietly, or never reported, will not appear here.

Hosted services are elsewhere. If your staff use AI tools in a browser rather than installing them, look at /tools/ instead.

How to use the list this week

1. Make a short inventory. Ask your developers or IT provider which of these 30 products your organization runs. Include tools on laptops, such as LM Studio or Jupyter, as well as anything on servers.
2. Open each product page. Start from /vulnerabilities/ and check the newest advisories against the version you run.
3. Follow the feeds. Add each product’s RSS feed to a feed reader that someone checks regularly. Our guide to subscribing to advisories by product walks through it.
4. Watch the site-wide view. The advisories page shows recent advisories across every product, which helps when you are not sure what is installed.
5. Revisit the inventory. When someone adds a new AI tool, check whether it is on this list and add its feed.

If you run a product with a high count, such as Langflow, n8n or Open WebUI, put its feed first. More advisories means more chances that an update you have not installed fixes something serious.