Cyber Threat Mitigation with Knowledge-Infused Reinforcement Learning and LLM-Guided Policies
Bibliographic record
Abstract
As cyber threats continue to evolve, there is a need for autonomous cyber defense (ACD) strategies capable of fast and context-aware responses. Reinforcement learning (RL) has shown promise for automating cyber defense by exploring and learning effective countermeasures, yet it often struggles with sparse reward signals and insufficient context to handle diverse attack scenarios. Furthermore, the convergence time taken by an RL agent is often high, which makes it difficult to train the RL agent in online settings. To address these challenges, we propose a large language model (LLM)-enhanced RL method that builds and queries a knowledge graph (KG) derived from agent-environment interactions. We leverage the pre-trained knowledge of an LLM on different cybersecurity frameworks and use the LLM to analyze a part of the KG to generate appropriate actions for the RL agent. We infuse the knowledge extracted from the LLM into the RL agent’s training loop in two ways. First, the state vector of the RL agent is augmented with the most effective action and its corresponding reward, as determined from the KG. Second, the suggested action from the LLM is used as a reference policy. In addition, we introduce a regularization term in the loss function to make the RL policy close to the reference policy. To validate our approach, we develop a custom RL environment guided by the MITRE ATT&CK framework, enabling the agent to generate tailored mitigation strategies for detected cyber attacks. Experimental results show that our proposed approach significantly outperforms the baseline RL by over $75 \%$ in terms of taking better mitigation actions.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".