diff --git a/malsim/agents/__init__.py b/malsim/agents/__init__.py
index 9b4bb0bb..830af82a 100644
--- a/malsim/agents/__init__.py
+++ b/malsim/agents/__init__.py
@@ -2,11 +2,16 @@
 from .passive_agent import PassiveAgent
 from .keyboard_input import KeyboardAgent
 from .searchers import BreadthFirstAttacker, DepthFirstAttacker
+from .heuristic_agent import (
+    DefendCompromisedDefender, DefendFutureCompromisedDefender
+)
 
 __all__ = [
     'PassiveAgent',
     'DecisionAgent',
     'KeyboardAgent',
     'BreadthFirstAttacker',
-    'DepthFirstAttacker'
+    'DepthFirstAttacker',
+    'DefendCompromisedDefender',
+    'DefendFutureCompromisedDefender'
 ]
diff --git a/malsim/agents/heuristic_agent.py b/malsim/agents/heuristic_agent.py
new file mode 100644
index 00000000..e1377207
--- /dev/null
+++ b/malsim/agents/heuristic_agent.py
@@ -0,0 +1,125 @@
+from __future__ import annotations
+from typing import Optional, TYPE_CHECKING
+import logging
+import math
+
+import numpy as np
+
+from .decision_agent import DecisionAgent
+
+if TYPE_CHECKING:
+    from maltoolbox.attackgraph import AttackGraphNode
+    from ..mal_simulator import MalSimAgentStateView
+
+logger = logging.getLogger(__name__)
+
+class DefendCompromisedDefender(DecisionAgent):
+    """A defender that defends compromised assets using notPresent"""
+
+    def __init__(self, agent_config, **_):
+        # Seed and rng not currently used
+        seed = (
+            agent_config["seed"]
+            if agent_config.get("seed")
+            else np.random.SeedSequence().entropy
+        )
+        self.rng = (
+            np.random.default_rng(seed)
+            if agent_config.get("randomize")
+            else None
+        )
+
+    def get_next_action(
+        self, agent_state: MalSimAgentStateView, **kwargs
+    ) -> Optional[AttackGraphNode]:
+
+        """Return an action that disables a compromised node"""
+
+        selected_node_cost = math.inf
+        selected_node = None
+
+        # To make it deterministic
+        possible_choices = list(agent_state.action_surface)
+        possible_choices.sort(key=lambda n: n.id)
+
+        for node in possible_choices:
+
+            if node.is_enabled_defense():
+                continue
+
+            node_cost = node.extras.get('reward', 0)
+
+            # Strategy:
+            # - Enabled the cheapest defense node
+            #   that has compromised child nodes
+            if node_cost < selected_node_cost:
+
+                node_has_compromised_child = (
+                    any(
+                        child_node.is_compromised()
+                        for child_node in node.children
+                    )
+                )
+
+                if node_has_compromised_child:
+                    selected_node = node
+                    selected_node_cost = node_cost
+
+        return selected_node
+
+
+class DefendFutureCompromisedDefender(DecisionAgent):
+    """A defender that defends compromised assets using notPresent"""
+
+    def __init__(self, agent_config, **_):
+        # Seed and rng not currently used
+        seed = (
+            agent_config["seed"]
+            if agent_config.get("seed")
+            else np.random.SeedSequence().entropy
+        )
+        self.rng = (
+            np.random.default_rng(seed)
+            if agent_config.get("randomize")
+            else None
+        )
+
+    def get_next_action(
+        self, agent_state: MalSimAgentStateView, **kwargs
+    ) -> Optional[AttackGraphNode]:
+
+        """Return an action that disables a compromised node"""
+
+        selected_node_cost = math.inf
+        selected_node = None
+
+        # To make it deterministic
+        possible_choices = list(agent_state.action_surface)
+        possible_choices.sort(key=lambda n: n.id)
+
+        for node in possible_choices:
+
+            if node.is_enabled_defense():
+                continue
+
+            node_cost = node.extras.get('reward', 0)
+
+            # Strategy:
+            # - Enabled the cheapest defense node
+            #   that has a non compromised child
+            #   that has a compromised parent.
+            if node_cost < selected_node_cost:
+
+                node_has_child_that_can_be_compromised = (
+                    any(
+                        any(p.is_compromised() for p in child_node.parents)
+                        and not child_node.is_compromised()
+                        for child_node in node.children
+                    )
+                )
+
+                if node_has_child_that_can_be_compromised:
+                    selected_node = node
+                    selected_node_cost = node_cost
+
+        return selected_node
diff --git a/malsim/scenario.py b/malsim/scenario.py
index f7612589..d7aa4f8e 100644
--- a/malsim/scenario.py
+++ b/malsim/scenario.py
@@ -22,7 +22,9 @@
     BreadthFirstAttacker,
     DepthFirstAttacker,
     KeyboardAgent,
-    PassiveAgent
+    PassiveAgent,
+    DefendCompromisedDefender,
+    DefendFutureCompromisedDefender
 )
 
 from .mal_simulator import AgentType, MalSimulator
@@ -32,6 +34,8 @@
     'BreadthFirstAttacker': BreadthFirstAttacker,
     'KeyboardAgent': KeyboardAgent,
     'PassiveAgent': PassiveAgent,
+    'DefendCompromisedDefender': DefendCompromisedDefender,
+    'DefendFutureCompromisedDefender': DefendFutureCompromisedDefender
 }
 
 deprecated_fields = [
diff --git a/tests/agents/test_heuristic_agents.py b/tests/agents/test_heuristic_agents.py
new file mode 100644
index 00000000..841e933b
--- /dev/null
+++ b/tests/agents/test_heuristic_agents.py
@@ -0,0 +1,131 @@
+from unittest.mock import MagicMock
+from maltoolbox.attackgraph import AttackGraphNode, Attacker
+from maltoolbox.language import LanguageGraph
+from malsim.mal_simulator import MalSimAgentStateView
+from malsim.agents import (
+    DefendCompromisedDefender,
+    DefendFutureCompromisedDefender
+)
+
+def test_defend_compromised_defender(dummy_lang_graph: LanguageGraph):
+    r"""
+            node1          node2
+            /    \         /   \
+        node3       node4        node5
+
+    """
+    dummy_or_attack_step = (
+        dummy_lang_graph.assets['DummyAsset']
+        .attack_steps['DummyOrAttackStep']
+    )
+
+    # Create nodes
+    node1 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=1)
+    node2 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=2)
+
+    node3 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=3)
+    node4 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=4)
+    node5 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=5)
+
+    # Connect nodes (Node1 -> Node2 -> Node3 -> Node4)
+    node1.children.add(node3)
+    node3.parents.add(node1)
+    node1.children.add(node4)
+    node4.parents.add(node1)
+
+    node2.children.add(node4)
+    node4.parents.add(node2)
+    node2.children.add(node5)
+    node5.parents.add(node5)
+
+    # Set up an attacker
+    attacker = Attacker(name="TestAttacker")
+    attacker.compromise(node4)
+
+    # Set up a mock MalSimAgentState
+    agent = MagicMock()
+    agent.action_surface = [node1, node2]
+
+    # Set up MalSimAgentStateView
+    agent_view = MalSimAgentStateView(agent)
+
+    # Configure BreadthFirstAttacker
+    agent_config = {"seed": 42, "randomize": False}
+    defender_ai = DefendCompromisedDefender(agent_config)
+
+    # Should pick cheapest one
+    node1.extras['reward'] = 100
+    node2.extras['reward'] = 10
+
+    # Get next action
+    action_node = defender_ai.get_next_action(agent_view)
+    assert action_node is not None, "Action node shouldn't be None"
+    assert action_node.id == node2.id
+
+    # Should pick cheapest one
+    node1.extras['reward'] = 10
+    node2.extras['reward'] = 100
+
+    # Get next action
+    action_node = defender_ai.get_next_action(agent_view)
+    assert action_node is not None, "Action node shouldn't be None"
+    assert action_node.id == node1.id
+
+
+def test_defend_future_compromised_defender(dummy_lang_graph: LanguageGraph):
+    r"""
+            node1              node2
+            /    \             /     \
+        node3      node4       |    node5
+                      |        |
+                       \      /
+                         node 6
+    """
+
+    dummy_or_attack_step = (
+        dummy_lang_graph.assets['DummyAsset']
+        .attack_steps['DummyOrAttackStep']
+    )
+
+    # Create nodes
+    node1 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=1)
+    node2 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=2)
+
+    node3 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=3)
+    node4 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=4)
+    node5 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=5)
+    node6 = AttackGraphNode(lg_attack_step=dummy_or_attack_step, node_id=6)
+
+    # Connect nodes (Node1 -> Node2 -> Node3 -> Node4)
+    node1.children.add(node3)
+    node3.parents.add(node1)
+    node1.children.add(node4)
+    node4.parents.add(node1)
+
+    node2.children.add(node6)
+    node6.parents.add(node2)
+    node2.children.add(node5)
+    node5.parents.add(node5)
+
+    node4.children.add(node6)
+    node6.parents.add(node4)
+
+    # Set up an attacker
+    attacker = Attacker(name="TestAttacker")
+    attacker.compromise(node4)
+
+    # Set up a mock MalSimAgentState
+    agent = MagicMock()
+    agent.action_surface = [node1, node2]
+
+    # Set up MalSimAgentStateView
+    agent_view = MalSimAgentStateView(agent)
+
+    # Configure BreadthFirstAttacker
+    agent_config = {"seed": 42, "randomize": False}
+    defender_ai = DefendFutureCompromisedDefender(agent_config)
+
+    # Should pick node 2 either way
+    action_node = defender_ai.get_next_action(agent_view)
+    assert action_node is not None, "Action node shouldn't be None"
+    assert action_node.id == node2.id