Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026Saarbrücken, Germany

Adaptive AI-Based Containment of Autonomous Cyber Attacks: A Reproducible Docker Cyber Range Study

Jayasankar Kumar Santhirani, Deivanai Shankar, Arjun Shetty, Karthik Venkatesh, Jesmilan Jesu · Team Saarlanders

Submitted to AI Incident Response Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Adaptive AI-Based Containment of Autonomous Cyber Attacks: A Reproducible Docker Cyber Range Study

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

The study evaluates whether an incident-history-reasoning defender outperforms a fixed response policy against an autonomous LLM attacker changing paths after containment. Using a minimal, isolated Docker cyber range featuring a legitimate workload, an unscripted attacker, and a policy engine enforcing deny-only actions, the research compares no defense, a stateful static playbook, and an AI defender across 40 measured episodes and four controls. The empirical results demonstrate that the static playbook prevented secret retrieval in all eight of its episodes, whereas the AI defender allowed retrieval in three of eight episodes with a fixed canary and four of eight with adaptive placement, while also inflicting greater availability losses on legitimate traffic. When the attacker adapted by switching routes post-block, both defenders successfully achieved containment in every instance, with the AI defender responding within a median of one to seven seconds compared to the playbook's twenty-one seconds. Ultimately, the study shows that in this environment, within-incident adaptation did not reduce overall attack success or impact compared to a fixed policy, leading to a proposed Adaptive Containment Requirement (ACR) emphasizing bounded authority and third-party verifiable conformance.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. - There are some unsubstantiated claims "Each containment action taken was a fixed, one-shot intervention, and the attacker adapted around it faster than a human loop could react"

    - the idea of adaptive containment is good and currently being researched, more investigations into different ways to implement this are welcome

    - It is hard to say conclusive things about this idea because there are so many ways for the offensive team to try and escape, as well as many ways for the defensive team to try and prevent this escape.

    - Significant LLM-written signs, this makes me less likely to read the full project as closely as I would otherwise

    - it's unmotivated and unclear why Gemini 3.7 Flash with low reasoning effort was used as the defender but GLM-5.3 (reasoning effort unspecified) was used as the attacker.

    - the experiment design and cyber range is quite complicated, making it difficult to figure out which are necessary parts and which are incidental (AI written?) complications.

    Read full reviewShow less

  2. What makes this approach compelling is how it pairs static playbooks and canary credentials with an LLM defender without over-privileging the model. Enforcing bounded authority—where the LLM proposes deny-only actions validated by an independent policy gate, backed by a deterministic playbook fallback—delivers fast, auditable containment via telemetry logs while keeping blast radiuses strictly contained. It strikes a pragmatic balance between adaptive reasoning and reliable, production-ready safety guardrails

Cite this project

@misc{santhirani2026adaptive,
  title = {{Adaptive AI-Based Containment of Autonomous Cyber Attacks: A Reproducible Docker Cyber Range Study}},
  author = {Jayasankar Kumar Santhirani and Deivanai Shankar and Arjun Shetty and Karthik Venkatesh and Jesmilan Jesu},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/adaptive-aibased-containment-of-autonomous-cyber-attacks-a-reproducible-docker-cyber-range-study-9kk3}},
  url = {https://apartresearch.com/sprints/projects/adaptive-aibased-containment-of-autonomous-cyber-attacks-a-reproducible-docker-cyber-range-study-9kk3}
}

More from AI Incident Response Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026