Read the Beforeitsnews.com story here. Advertise at Before It's News here.
Profile image
By Security Warrior (Reporter)
Contributor profile | More stories
Story Views
Now:
Last hour:
Last 24 hours:
Total:

WhatDR Meets AI: Is “AIDR” the Next EDR, or Three Racoons in a Trench Coat?

% of readers think this story is Fact. Add your two cents.


Warning: this blog is another experiment. The idea for a blog came from a thought that has been bugging me for months. The content is researched, debated, refined, “hardened” by a Gemini skill I created (using Gemini 4 Argon). Then Claude Fable 5.2 performed a peer review, and refined the text. I then reviewed it and utterly hated it! It went into the trash. I refined the idea seed that I feed into my Gemini skill and tried again. The whole process ran better this time. And then I lightly edited it, and tuned a few jokes. I replaced 3 kids with 3 racoons (racoons are funnier!). Let me know if this sucks, and I will stop. Promise!

Back in March 2024, in WhatDR or What Detection Domain Needs Its Own Tools?, I begged the universe: “And, gods, please, no AIDR…” Naturally, cybersecurity marketing gods did not listen. I suspect they read my plea as a product spec. (Reminder: I have been out of the acronym-crafting business since 2019, so do not blame me for this one.)

By late 2026 the pitch is everywhere: agents are the new endpoint, this is “the EDR moment for AI”, and you need a dedicated AI Detection and Response (AIDR) product. Right? This is no longer only crazy startup talk. CrowdStrike says it is defining the AIDR category and draws the EDR comparison itself. Meanwhile, the grumpy minimalist in the back row mutters: “Relax, AI is just JSON over HTTPS. Ship the logs to SIEM and go home.”

Both camps are half-wrong. Well, more like two-thirds wrong, but anyhow.

My short version: AI agents need runtime detection and response, and my own 2024 “WhatDR” test says so. What they do not need is one more new console, because what is sold as “AIDR” today is three different jobs inside one trench coat. One is new, one is not detection and response at all, and one already belongs to your SIEM.

Three Raccoons in the AIDR Trench Coat

When I named EDR in 2013, it earned its seat in the SOC because the endpoint is a coherent boundary. Endpoint threats detection could not be reduced to OS logs (“Sysmon into SIEM” was always a sad, poor person’s EDR, or a decent “second EDR”), and one local sensor could watch memory, process trees, DLLs and kernel activity, then isolate the host within seconds. It can both sense and act.

Now look at typical 2026 “AIDR” pitch. I see 3 raccoons in a trenchcoat model here.


a 3 racoon model of AIDR

There are three jobs, and the easiest way to tell them apart is to ask what each one watches:

  • Top raccoon watches the agents you run: the inline agent and model gate. Inspecting prompts, retrieved context and Model Context Protocol (MCP) tool calls mid-flight, and blocking an injected instruction or an unauthorized tool call before it executes. As I argued recently, the model has preferences, but the harness got controls. Architecturally this is close to RASP (now dead and forgotten) plus an API gateway: heavy on prevention, light on investigation. This is the new job. IMHO, very real.
  • Middle raccoon watches humans using AI: workforce AI control. Catching the seven shades of Shadow AI: unsanctioned browser extensions, local coding agents, employees pasting M&A spreadsheets into public models. This is useful, but it is not detection and response, in my book. Most of the time there is no attacker and no incident, only a policy to enforce. It is a hall monitor in a detective’s coat, and your DLP, CASB and EDR budgets already pay for it. (In that WhatDR post I said DDR probably should not exist because data is not a technology domain. Same logic, new sticker.)
  • Bottom raccoon watches an agent’s trail across your other systems: cross-domain correlation. Catching an agent whose OAuth token reads a Jira ticket, queries a customer table and posts a summary to an external Slack channel. That is application visibility from the SOC Visibility Quad joined with identity and data logs. This one is real detection and response, and it already has an owner: SIEM.

(Some decks also file “catching AI-generated phishing” under AIDR. Yuck! That belongs in the same hall of shame as “VMDR.”)

Running AI Through the WhatDR Test

The original post had seven decision dimensions for when a domain needs its own “?DR” or “*DR” tool versus just using a broad tool like SIEM. BTW, there and here as well, this was a lot more about D, detection, than about R, response.

Here are the seven, scored for AI agents, with the raccoon each row belongs to.


AI scored vs WhatDR critia + racoons

Read the “2024 verdict” column honestly: rows 2, 3, 5 and 6 point to a dedicated tool, and all four belong to the top raccoon. So by my own 2024 test, the top raccoon passes and this means that AIDR should exist. Agent runtime detection and response is a real domain with its own telemetry, its own context and its own response actions. Somebody has to build that capability.

The middle raccoon shows up only in row 4, and it should not be taking this test at all. WhatDR was a test for detection domains, and policy enforcement on employee AI use is not one. The bottom raccoon is row 7, which was always SIEM.

So why not hand the top raccoon a console and call it the next EDR?

I made two arguments (and then got tired of it, but I think there are more):

  • Inherent broad needs. EDR got its console in an era when the SOC had nowhere else to investigate an endpoint. An agent incident is cross-domain from the first minute, because the question that matters is what the agent’s identity touched in other systems. “AIDR console” would be too narrow to be useful, or will have to become SIEM to work.
  • RASP history. I compared the gate to RASP on purpose. RASP had to live inside somebody else’s application, and that is a big part of why it ended up as a feature and then a bad memory. The agent gate has the same dependency on the harness, so I expect it to ship with the harness, the AI gateway or the platform.
  • You do not own most harnesses. The agents inside your SaaS suites run in the vendor’s harness, and you cannot put a sensor there. You get whatever audit logs and admin controls the vendor exposes, and those go to SIEM. For many lean teams this is the majority of their agents.

My verdict: separate sensor — yes, separate control point — maybe, separate console — likely no. Note that I have changed my mind on this kind of thing before. In 2022, I leaned toward CDR deserving its own category. By WhatDR time (2024) it sat in my gray zone, and my read is that much of it has since ended up as a capability inside CNAPP and SIEM. In 2026, there are not many CDR vendors left, if any. I expect AIDR to take the same road, but faster (because AI).

When would I be wrong? If you build and run many of your own agents, own every harness, and have a dedicated AI security team, a separate console for that team can be justified. That describes a small number of AI-heavy companies, and they know who they are.

Why “Just Ship It to SIEM” Fails

Saying “AI is just another log source” in 2026 is as naive as saying “Sysmon into SIEM is as good as EDR” was in 2014. (I am a constructive cynic, not a full-on SIEM museum curator.)

I see four reasons, but, come on, this is kinda obvious:

  • Data becomes control flow. Application logs assume deterministic code paths. When an agent reads an untrusted email or a poisoned wiki page, the content can redirect what the agent does next. AgentDojo built a benchmark of 97 tasks and 629 security test cases around exactly this. A log line written after the tool call cannot make the allow-or-block decision. It has to be made at the harness, before the call runs. A key argument for “?DR” is a need for control on the spot.
  • Timing. A SIEM rule that runs every few minutes will notice that an agent read an untrusted document and then called an export API. By then the export has happened. Detection after the fact is still worth having, but it does not help much for rapid sequence of automated actions.
  • Response granularity. If the only lever your SOC has is revoking the shared service-account token, containment means an outage for every legitimate user of that application. You need to be able to end one agent session or block one tool call.
  • Sensitivity, then volume. An illustrative example with round numbers (yours will differ): 500 agents running 200 tasks a day is 100,000 runs. If each run has 10 steps carrying about 50 KB of prompt history, retrieved documents and tool output, that is 500 KB per run, or about 50 GB a day of unstructured text. This is fine, for now. Still, having the harness evaluate locally and emit only structured and useful logs, analyzed on the spot is better AND cheaper.

Potentially Useful Actions Based on This

This was fun — we concluded the world does need AIDR, but it may not be what people call that name.

To reduce confusion, do these:

  1. Open the trench coat on every “AIDR” pitch. Ask which raccoon the vendor builds. Score the gate against the seven rows above. Judge workforce AI control as a policy tool against the DLP and CASB you already own (this isn’t AIDR). Evaluate cross-domain correlation against your SIEM collect and detect capabilities.
  2. Inventory the agents whose harness you do not own. For each SaaS agent, find out which audit logs and kill switches the vendor exposes, and get those logs into SIEM. Here you do need SIEM, not AIDR.
  3. Add an AI-telemetry question to your next MDR or SIEM review. Ask whether their detection content uses gen_ai tool-call telemetry and correlates it with identity and data-access logs. No tool removes the learning cost of knowing which data an agent is allowed to touch.
  4. Keep raw prompts out of SIEM by default. Follow an Output-Driven SIEM approach: have gateways and harnesses emit structured OpenTelemetry spans with a conversation ID, tool names and guardrail verdicts, and keep redacted payloads in local trace storage for on-demand retrieval.

As much harder extra credit:

  1. Test your “R” at the session level. Run an API or Die audit on your agent orchestrators and codify Context as Code for agent identities. If your SOC or MDR cannot end a single agent session, remove a single tool permission, or quarantine a poisoned document without revoking the application’s master key, you have no usable response.

So, Is AIDR the Next EDR?

One raccoon of three is, and what it gives you is “the next EDR”, perhaps. That is also the math behind “two-thirds wrong.” Some of the AIDR fans are wrong about two raccoons of three.

Every new computing substrate replays the same sequence. First we pretend the old log pipelines are enough. Then somebody sells a new console for the new domain. Then operations forces a settlement: sensors and inline controls live in the new runtime, and cross-domain detection and investigation stay in SIEM. AI is running that sequence faster than cloud did.

Just do not give the trench coat its own island in your SOC.

Agree? Disagree?

[A.C. — note to self: this is v7.5 Gemini skill using 4 Argon, take 2]

Related blogs:



WhatDR Meets AI: Is “AIDR” the Next EDR, or Three Racoons in a Trench Coat? was originally published in Anton on Security on Medium, where people are continuing the conversation by highlighting and responding to this story.


Originally published at Medium.

About me: http://www.chuvakin.org

This blog focuses on SIEM, log management, PCI DSS compliance and other information security issues. Check out more articles like this here: http://chuvakin.blogspot.com/


Source: http://chuvakin.blogspot.com/2026/11/whatdr-meets-ai-is-aidr-next-edr-or.html


Before It’s News® is a community of individuals who report on what’s going on around them, from all around the world.

Anyone can join.
Anyone can contribute.
Anyone can become informed about their world.

"United We Stand" Click Here To Create Your Personal Citizen Journalist Account Today, Be Sure To Invite Your Friends.

Before It’s News® is a community of individuals who report on what’s going on around them, from all around the world. Anyone can join. Anyone can contribute. Anyone can become informed about their world. "United We Stand" Click Here To Create Your Personal Citizen Journalist Account Today, Be Sure To Invite Your Friends.


LION'S MANE PRODUCT


Try Our Lion’s Mane WHOLE MIND Nootropic Blend 60 Capsules


Mushrooms are having a moment. One fabulous fungus in particular, lion’s mane, may help improve memory, depression and anxiety symptoms. They are also an excellent source of nutrients that show promise as a therapy for dementia, and other neurodegenerative diseases. If you’re living with anxiety or depression, you may be curious about all the therapy options out there — including the natural ones.Our Lion’s Mane WHOLE MIND Nootropic Blend has been formulated to utilize the potency of Lion’s mane but also include the benefits of four other Highly Beneficial Mushrooms. Synergistically, they work together to Build your health through improving cognitive function and immunity regardless of your age. Our Nootropic not only improves your Cognitive Function and Activates your Immune System, but it benefits growth of Essential Gut Flora, further enhancing your Vitality.



Our Formula includes: Lion’s Mane Mushrooms which Increase Brain Power through nerve growth, lessen anxiety, reduce depression, and improve concentration. Its an excellent adaptogen, promotes sleep and improves immunity. Shiitake Mushrooms which Fight cancer cells and infectious disease, boost the immune system, promotes brain function, and serves as a source of B vitamins. Maitake Mushrooms which regulate blood sugar levels of diabetics, reduce hypertension and boosts the immune system. Reishi Mushrooms which Fight inflammation, liver disease, fatigue, tumor growth and cancer. They Improve skin disorders and soothes digestive problems, stomach ulcers and leaky gut syndrome. Chaga Mushrooms which have anti-aging effects, boost immune function, improve stamina and athletic performance, even act as a natural aphrodisiac, fighting diabetes and improving liver function. Try Our Lion’s Mane WHOLE MIND Nootropic Blend 60 Capsules Today. Be 100% Satisfied or Receive a Full Money Back Guarantee. Order Yours Today by Following This Link.


Report abuse

Comments

Your Comments
Question   Razz  Sad   Evil  Exclaim  Smile  Redface  Biggrin  Surprised  Eek   Confused   Cool  LOL   Mad   Twisted  Rolleyes   Wink  Idea  Arrow  Neutral  Cry   Mr. Green

MOST RECENT
Load more ...

SignUp

Login