AI Security editorial
AI Caramba — landing
Short reads on AI Security — foundations primers, situational self-awareness, and where AI meets data paths. Govern, protect, and manage as a reading frame. Pieces labeled Draft / PROMISE are editorial honesty, not final copy. English primary.
When models decide how to work: Self-awareness, resources, and the limits of the frontier
Frontier models increasingly choose how they work — self-do, fan-out, rewrite standards. Limited situational awareness makes that a control problem, not a feature footnote.
Latest
Draft / PROMISE · AI Foundations · AI × Data Security
-
What self-awareness means for multi-engine setups and control planes — and why T3 Code suddenly makes senseDraft
Multi-engine means heterogeneous models as tools. A control plane holds threads, approvals, and diffs — T3 Code is one verified example.
-
What “AI Security” actually means (and what it does not)Draft
AI Security is how models, tools, and data paths fail — see, do, remember — not a vendor checklist or suite pitch.
-
Models, tools, and agents: three layers readers mix upDraft
A model answers. A tool reaches outward. An agent sequences both. Name the layer before you argue about controls.
-
Prompts are not policy: where intent stops and control startsDraft
A prompt states intent. Policy is what systems enforce when intent fails — AuthZ, DLP, approvals, audit.
-
Training data, inference data, and logs: three different risksDraft
What entered the model is not what enters a session, and neither is what persists after. Controls differ.
-
When an AI tool becomes a data pathDraft
The moment an assistant can read mail, files, or tickets, it is no longer only a model. It is a data path with a blast radius.
-
Putting business data into a prompt: a practical risk checklistDraft
A prompt is a shipping label. Before you paste the spreadsheet, know what you are sending, who can see the transit, and what outlives the answer.
-
Retrieval and RAG: useful, leaky if unboundedDraft
Retrieval makes answers better by fetching context. Unbounded retrieval makes the model a search engine over everything it should never have seen.
-
Agents that can act: blast radius before autonomy theaterDraft
An agent that sequences tools is a permission graph with a loop. Map blast radius before you celebrate autonomy.
-
AI Security as a reading path: foundations → data → intersectionDraft
AI Security is not a pile of posts. It is a path: how systems fail, what data is at stake, and where the two meet.