Your SOC runs on what your analysts never wrote down.

Close to half the alerts a SOC opens are work the client had already approved, and an analyst closes each one because she knows the client. Tacit sits between your SIEM and your desk, captures that reasoning from the decisions your analysts already make, and closes the next one with the reason on the record. It is a model we built ourselves and not a language model, so it is cheap enough to read every alert and you can see why it made each call.

See CyberSOC
  1. 38%

    of the whole queue closed on public data, which is 84% of the alerts that were expected work.

  2. 42

    escalations found in 1,000 alerts in 90 seconds, in a proof of concept that is still running.

Our brief

The longer read: what a knowledge creating company is, and why the security operations centre is where ours starts.

Knowledge
creating
company.

Tacit captures the judgement behind the decisions a team already makes, and puts it to work. At the centre is a judgement model of our own, which reads the decisions a desk has made, learns how that desk decides, and shows where one person’s habit departs from it. The first desk we have put it on is the security operations centre.

The Tacit platform

Built to make
the expert judgement
machine-legible.

§ 01 / Thesis

Thesis, briefly.

AI agents are good at retrieval. They are good at language. In regulated work they fail at the one thing that matters most: judgement.

The published evidence on enterprise AI is now consistent enough to read as a pattern. BCG and McKinsey report 88% adoption against 5% meaningful returns. Gartner reports 85% of projects never reach production. MIT's 2024 study found 95% of enterprise generative AI pilots produced no measurable ROI. The standard explanation is that the models are immature, that retrieval needs to be better — that explanation has been wrong for at least two years.

The agents are failing because the AI industry has been trying to capture expertise as content. Most of the meta decisions that matter have never been written down. The part of expertise that defines the senior expert is not knowledge that can be made larger or more accessible. It is a runtime construct, a live cognitive process that decides, under uncertainty, which heuristic fires. We make that runtime observable, persistent, and queryable for the agents that will inherit the work.

We start in the security operations centre. Close to half the alerts an analyst opens there are work the client had already approved, and what closes them is something the analyst knows and the SIEM does not. That is judgement nobody wrote down, in a place where it is used hundreds of times a day, and it is where we begin.

§ 02 / Diagnosis

The shape of
the problem.

  1. Tacit knowledge does not exist on paper. The documents an LLM can read describe the world after the experts have already left the room.

  2. A senior SOC analyst cannot tell you, in words, why she closed an alert in four seconds. She knows the client, so she closed it.

  3. A senior radiologist sees the malignancy before they can articulate why. The seeing is the expertise. The words are a post-hoc reconstruction.

  4. A control-room operator detects a drift through a posture, a hum, a feel. The SCADA log shows nothing yet.

  5. These signals do not survive into training data. So they do not survive into the model. So they do not survive into the agent.

  6. What survives is the documentation: the safety bulletins, the discharge summaries, the runbooks — the thin paper trail of decisions, not the decisions themselves.

  7. LLMs trained on that paper trail are fluent but ungrounded. They are plausible at the centre of the distribution and dangerous at the margins.

  8. The margins are where money, lives and licenses live. Which is exactly where regulated work spends its time.

§ 03 / Engine

What we are building.

01 — Capture

From the work itself

Tacit reads the decisions a desk has already made and what a client says in the course of the work: a message, a ticket, a change window. Nobody is interviewed, and an analyst never has to type an explanation.

02 — Structure

A register of reasons

Every reason is kept as a card: what it covers, who vouched for it, and the day it lapses. After that day the alert goes back to a person, so nothing is excused for good.

03 — Decide

A judgement model

A small statistical model of our own, not a language model. It learns what moves the desk as a whole and where one analyst’s habit departs from it, it is cheap and fast enough to read every alert, and you can read it one answer at a time.

04 — Audit

A ledger you can check

Every close is an entry sealed to the one before, with its reason attached. A client or an auditor can check that a record is whole without signing in and without taking our word for it.

§ 04 / Timing

Why now.

01

A small model can now do what a language model was too slow for.

A language model is too slow and too costly to sit on every decision a desk makes. A small statistical model trained on that desk’s own decisions is neither. On Microsoft’s public GUIDE data ours closes 84% of benign positives, at a setting where 1 in 100 true positives is closed as well.

02

A close now needs a reason somebody can check.

The EU AI Act is coming into force in stages, and clients and auditors have started asking the second question: why was that closed? A score cannot answer it. A reason on the record, with who vouched for it and when it lapses, can.

03

The people who know keep leaving.

An analyst moves on, or the client changes provider, and what they knew about that client goes with them. The desk starts again each time, and the same alert is argued over from the beginning.

04

AI triage is arriving whether the desk asked for it or not.

Vendors and OEMs now ship AI triage inside the tools a SOC already pays for, and the firm still answers when the AI gets a call wrong. Most of the market checks one language model with another. Checking an agent against how the desk itself decides is our second product.

§ 05 / Markets

Where we work.

CyberSOC · for SOC teams and the people who run them

Capturing how a SOC’s analysts reason.

Tacit sits between the SIEM and the desk. It closes what the client already explained with the reason attached, finds the alerts that need an L2 within seconds, and brings a new analyst up against the desk’s own decisions.

It starts read-only, with a two-week review of the decisions the desk has already made, and it goes live one kind of alert at a time and only where the desk says so. We sell to managed security providers and to firms that run their own SOC.

Design partnersFrom December 2026
First platformsSentinel & Defender
Quality assurance · in the lab

An independent check on the AI agents your vendors ship.

The same judgement model checks the calls an AI agent has already made against what actually happened, and against how your own team decides, and shows where the evidence supports letting it act alone.

A vendor cannot do that independently, and most of the market does it by asking one language model to grade another. It is not a product yet, and it starts in the SOC because that is where we already know how the desk decides.

PhaseIn the lab
Starts withSecurity operations

— The knowledge creating company —

Every decision runs on judgement. We recover it, and show how far you can rely on it

Talk to us Read the full thesis