diff --git a/AnomalyDetection/LibLinear/src/main/java/org/tribuo/anomaly/liblinear/LibLinearAnomalyTrainer.java b/AnomalyDetection/LibLinear/src/main/java/org/tribuo/anomaly/liblinear/LibLinearAnomalyTrainer.java
index 5caf5a719..7efdd71da 100644
--- a/AnomalyDetection/LibLinear/src/main/java/org/tribuo/anomaly/liblinear/LibLinearAnomalyTrainer.java
+++ b/AnomalyDetection/LibLinear/src/main/java/org/tribuo/anomaly/liblinear/LibLinearAnomalyTrainer.java
@@ -41,7 +41,11 @@
 
 /**
  * A {@link Trainer} which wraps a liblinear-java anomaly detection trainer using a one-class SVM.
- *
+ * <p>
+ * Note the train method is synchronized on {@code LibLinearTrainer.class} due to a global RNG in liblinear-java.
+ * This is insufficient to ensure reproducibility if liblinear-java is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Fan RE, Chang KW, Hsieh CJ, Wang XR, Lin CJ.
@@ -127,6 +131,9 @@ protected List<Model> trainModels(Parameter curParams, int numFeatures, FeatureN
         data.x = features;
         data.n = numFeatures;
 
+        // Note this isn't sufficient for reproducibility as it doesn't cope with concurrency.
+        // Concurrency safety is handled by the global lock on LibLinearTrainer.class in LibLinearTrainer.train.
+        Linear.resetRandom();
         return Collections.singletonList(Linear.train(data,curParams));
     }
 
diff --git a/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/LibSVMAnomalyTrainer.java b/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/LibSVMAnomalyTrainer.java
index 17227476b..7e0982716 100644
--- a/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/LibSVMAnomalyTrainer.java
+++ b/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/LibSVMAnomalyTrainer.java
@@ -22,6 +22,7 @@
 import org.tribuo.Example;
 import org.tribuo.ImmutableFeatureMap;
 import org.tribuo.ImmutableOutputInfo;
+import org.tribuo.Trainer;
 import org.tribuo.anomaly.Event;
 import org.tribuo.anomaly.Event.EventType;
 import org.tribuo.common.libsvm.LibSVMModel;
@@ -38,11 +39,16 @@
 import java.util.Collections;
 import java.util.List;
 import java.util.Map;
+import java.util.SplittableRandom;
 import java.util.logging.Logger;
 
 /**
  * A trainer for anomaly models that uses LibSVM.
  * <p>
+ * Note the train method is synchronized on {@code LibSVMTrainer.class} due to a global RNG in LibSVM.
+ * This is insufficient to ensure reproducibility if LibSVM is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Chang CC, Lin CJ.
@@ -66,11 +72,20 @@ public class LibSVMAnomalyTrainer extends LibSVMTrainer<Event> {
     protected LibSVMAnomalyTrainer() {}
 
     /**
-     * Creates a one-class LibSVM trainer using the supplied parameters.
-     * @param parameters The training parameters.
+     * Creates a one-class LibSVM trainer using the supplied parameters and {@link Trainer#DEFAULT_SEED}.
+     * @param parameters The SVM training parameters.
      */
     public LibSVMAnomalyTrainer(SVMParameters<Event> parameters) {
-        super(parameters);
+        this(parameters, Trainer.DEFAULT_SEED);
+    }
+
+    /**
+     * Creates a one-class LibSVM trainer using the supplied parameters and RNG seed.
+     * @param parameters The SVM parameters.
+     * @param seed The RNG seed for LibSVM's internal RNG.
+     */
+    public LibSVMAnomalyTrainer(SVMParameters<Event> parameters, long seed) {
+        super(parameters,seed);
     }
 
     /**
@@ -100,7 +115,7 @@ protected LibSVMModel<Event> createModel(ModelProvenance provenance, ImmutableFe
     }
 
     @Override
-    protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs) {
+    protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs, SplittableRandom localRNG) {
         svm_problem problem = new svm_problem();
         problem.l = outputs[0].length;
         problem.x = features;
@@ -112,6 +127,9 @@ protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures,
         if(checkString != null) {
             throw new IllegalArgumentException("Error checking SVM parameters: " + checkString);
         }
+        // This is safe because we synchronize on LibSVMTrainer.class in the train method to
+        // ensure there is no concurrent use of the rng.
+        svm.rand.setSeed(localRNG.nextLong());
         return Collections.singletonList(svm.svm_train(problem, curParams));
     }
 
diff --git a/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/SVMAnomalyType.java b/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/SVMAnomalyType.java
index cbf45e5a0..e80261470 100644
--- a/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/SVMAnomalyType.java
+++ b/AnomalyDetection/LibSVM/src/main/java/org/tribuo/anomaly/libsvm/SVMAnomalyType.java
@@ -64,7 +64,7 @@ public SVMAnomalyType(SVMMode type) {
 
     @Override
     public boolean isClassification() {
-        return true;
+        return false;
     }
 
     @Override
@@ -74,7 +74,7 @@ public boolean isRegression() {
 
     @Override
     public boolean isAnomaly() {
-        return false;
+        return true;
     }
 
     @Override
diff --git a/Classification/LibLinear/src/main/java/org/tribuo/classification/liblinear/LibLinearClassificationTrainer.java b/Classification/LibLinear/src/main/java/org/tribuo/classification/liblinear/LibLinearClassificationTrainer.java
index d3583e9f8..0940a1817 100644
--- a/Classification/LibLinear/src/main/java/org/tribuo/classification/liblinear/LibLinearClassificationTrainer.java
+++ b/Classification/LibLinear/src/main/java/org/tribuo/classification/liblinear/LibLinearClassificationTrainer.java
@@ -44,7 +44,11 @@
 
 /**
  * A {@link Trainer} which wraps a liblinear-java classifier trainer.
- *
+ * <p>
+ * Note the train method is synchronized on {@code LibLinearTrainer.class} due to a global RNG in liblinear-java.
+ * This is insufficient to ensure reproducibility if liblinear-java is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Fan RE, Chang KW, Hsieh CJ, Wang XR, Lin CJ.
@@ -114,6 +118,9 @@ protected List<Model> trainModels(Parameter curParams, int numFeatures, FeatureN
         data.n = numFeatures;
         data.bias = 1.0;
 
+        // Note this isn't sufficient for reproducibility as it doesn't cope with concurrency.
+        // Concurrency safety is handled by the global lock on LibLinearTrainer.class in LibLinearTrainer.train.
+        Linear.resetRandom();
         return Collections.singletonList(Linear.train(data,curParams));
     }
 
diff --git a/Classification/LibLinear/src/test/java/org/tribuo/classification/liblinear/TestLibLinearModel.java b/Classification/LibLinear/src/test/java/org/tribuo/classification/liblinear/TestLibLinearModel.java
index 5c9277b3a..9272585df 100644
--- a/Classification/LibLinear/src/test/java/org/tribuo/classification/liblinear/TestLibLinearModel.java
+++ b/Classification/LibLinear/src/test/java/org/tribuo/classification/liblinear/TestLibLinearModel.java
@@ -112,7 +112,6 @@ public void testMulticlass() throws IOException, ClassNotFoundException {
         for (Example<Label> example : examples) {
             model.getExcuse(example);
         }
-        //System.out.println("*** PASSED: " + prefix);
     }
 
     private void checkModelType(LinearType modelType) throws IOException, ClassNotFoundException {
@@ -129,7 +128,6 @@ private void checkModelType(LinearType modelType) throws IOException, ClassNotFo
         for (Example<Label> example : examples) {
             model.getExcuse(example);
         }
-        //System.out.println("*** PASSED: " + prefix);
     }
 
 
@@ -184,6 +182,19 @@ public Model<Label> testLibLinear(Pair<Dataset<Label>,Dataset<Label>> p) {
         return m;
     }
 
+    @Test
+    public void testReproducible() {
+        // Note this test will need to change if LibLinearTrainer grows a per Problem RNG.
+        Pair<Dataset<Label>,Dataset<Label>> p = LabelledDataGenerator.denseTrainTest();
+        Model<Label> m = t.train(p.getA());
+        Map<String, List<Pair<String,Double>>> mFeatures = m.getTopFeatures(-1);
+
+        Model<Label> mTwo = t.train(p.getA());
+        Map<String, List<Pair<String,Double>>> mTwoFeatures = mTwo.getTopFeatures(-1);
+
+        assertEquals(mFeatures,mTwoFeatures);
+    }
+
     @Test
     public void testDenseData() {
         Pair<Dataset<Label>,Dataset<Label>> p = LabelledDataGenerator.denseTrainTest();
diff --git a/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationModel.java b/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationModel.java
index 3be699c34..dbbb3c0a5 100644
--- a/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationModel.java
+++ b/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationModel.java
@@ -92,6 +92,10 @@ public class LibSVMClassificationModel extends LibSVMModel<Label> {
         }
     }
 
+    /**
+     * Returns the number of support vectors.
+     * @return The number of support vectors.
+     */
     public int getNumberOfSupportVectors() {
         return models.get(0).SV.length;
     }
diff --git a/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationTrainer.java b/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationTrainer.java
index 26d399722..e1de14fb7 100644
--- a/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationTrainer.java
+++ b/Classification/LibSVM/src/main/java/org/tribuo/classification/libsvm/LibSVMClassificationTrainer.java
@@ -22,6 +22,7 @@
 import org.tribuo.Example;
 import org.tribuo.ImmutableFeatureMap;
 import org.tribuo.ImmutableOutputInfo;
+import org.tribuo.Trainer;
 import org.tribuo.classification.Label;
 import org.tribuo.classification.WeightedLabels;
 import org.tribuo.common.libsvm.LibSVMModel;
@@ -39,11 +40,16 @@
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.SplittableRandom;
 import java.util.logging.Logger;
 
 /**
  * A trainer for classification models that uses LibSVM.
  * <p>
+ * Note the train method is synchronized on {@code LibSVMTrainer.class} due to a global RNG in LibSVM.
+ * This is insufficient to ensure reproducibility if LibSVM is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Chang CC, Lin CJ.
@@ -69,10 +75,27 @@ public class LibSVMClassificationTrainer extends LibSVMTrainer<Label> implements
     @Config(description="Use Label specific weights.")
     private Map<String,Float> labelWeights = Collections.emptyMap();
 
+    /**
+     * For OLCUT.
+     */
     protected LibSVMClassificationTrainer() {}
 
+    /**
+     * Constructs a classification LibSVM trainer using the specified parameters
+     * and {@link Trainer#DEFAULT_SEED}.
+     * @param parameters The SVM parameters.
+     */
     public LibSVMClassificationTrainer(SVMParameters<Label> parameters) {
-        super(parameters);
+        this(parameters, Trainer.DEFAULT_SEED);
+    }
+
+    /**
+     * Constructs a classification LibSVM trainer using the specified parameters and seed.
+     * @param parameters The SVM parameters.
+     * @param seed The RNG seed for LibSVM's internal RNG.
+     */
+    public LibSVMClassificationTrainer(SVMParameters<Label> parameters, long seed) {
+        super(parameters,seed);
     }
 
     /**
@@ -92,7 +115,7 @@ protected LibSVMModel<Label> createModel(ModelProvenance provenance, ImmutableFe
     }
 
     @Override
-    protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs) {
+    protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs, SplittableRandom localRNG) {
         svm_problem problem = new svm_problem();
         problem.l = outputs[0].length;
         problem.x = features;
@@ -104,6 +127,9 @@ protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures,
         if(checkString != null) {
             throw new IllegalArgumentException("Error checking SVM parameters: " + checkString);
         }
+        // This is safe because we synchronize on LibSVMTrainer.class in the train method to
+        // ensure there is no concurrent use of the rng.
+        svm.rand.setSeed(localRNG.nextLong());
         return Collections.singletonList(svm.svm_train(problem, curParams));
     }
 
diff --git a/Classification/LibSVM/src/test/java/org/tribuo/classification/libsvm/TestLibSVM.java b/Classification/LibSVM/src/test/java/org/tribuo/classification/libsvm/TestLibSVM.java
index 557120620..651ff1be5 100644
--- a/Classification/LibSVM/src/test/java/org/tribuo/classification/libsvm/TestLibSVM.java
+++ b/Classification/LibSVM/src/test/java/org/tribuo/classification/libsvm/TestLibSVM.java
@@ -17,6 +17,7 @@
 package org.tribuo.classification.libsvm;
 
 import com.oracle.labs.mlrg.olcut.util.Pair;
+import libsvm.svm_model;
 import org.tribuo.CategoricalIDInfo;
 import org.tribuo.CategoricalInfo;
 import org.tribuo.Dataset;
@@ -56,12 +57,14 @@
 import java.io.ObjectInputStream;
 import java.nio.file.Path;
 import java.nio.file.Paths;
+import java.util.Arrays;
 import java.util.List;
 import java.util.Locale;
 import java.util.Map;
 
 import static org.junit.jupiter.api.Assertions.assertArrayEquals;
 import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
 import static org.junit.jupiter.api.Assertions.assertNotNull;
 import static org.junit.jupiter.api.Assertions.assertThrows;
 import static org.junit.jupiter.api.Assertions.assertTrue;
@@ -186,6 +189,42 @@ public Model<Label> testLibSVM(Pair<Dataset<Label>,Dataset<Label>> p) {
         return m;
     }
 
+    @Test
+    public void testReproducibility() {
+        Pair<Dataset<Label>,Dataset<Label>> p = LabelledDataGenerator.denseTrainTest();
+        long seed = 42L;
+        SVMParameters<Label> params = new SVMParameters<>(new SVMClassificationType(SVMMode.NU_SVC),KernelType.RBF);
+        params.setProbability();
+        LibSVMTrainer<Label> first = new LibSVMClassificationTrainer(params,seed);
+        LibSVMModel<Label> firstModel = first.train(p.getA());
+
+        LibSVMTrainer<Label> second = new LibSVMClassificationTrainer(params,seed);
+        LibSVMModel<Label> secondModel = second.train(p.getA());
+
+        LibSVMModel<Label> thirdModel = first.train(p.getA());
+        LibSVMModel<Label> fourthModel = second.train(p.getA());
+
+        svm_model m = firstModel.getInnerModels().get(0);
+        svm_model mTwo = secondModel.getInnerModels().get(0);
+        svm_model mThre = thirdModel.getInnerModels().get(0);
+        svm_model mFour = fourthModel.getInnerModels().get(0);
+
+        // One and two use the same RNG seed and should be identical
+        assertArrayEquals(m.sv_coef,mTwo.sv_coef);
+        assertArrayEquals(m.probA,mTwo.probA);
+        assertArrayEquals(m.probB,mTwo.probB);
+
+        // The RNG state of three has diverged and should produce a different model.
+        assertFalse(Arrays.equals(mTwo.probA,mFour.probA));
+        assertFalse(Arrays.equals(mTwo.probB,mFour.probB));
+
+        // The RNG state for three and four are the same so the two models should be the same.
+        assertArrayEquals(mFour.sv_coef,mThre.sv_coef);
+        assertArrayEquals(mFour.probA,mThre.probA);
+        assertArrayEquals(mFour.probB,mThre.probB);
+
+    }
+
     @Test
     public void testDenseData() {
         Pair<Dataset<Label>,Dataset<Label>> p = LabelledDataGenerator.denseTrainTest();
diff --git a/Common/LibLinear/src/main/java/org/tribuo/common/liblinear/LibLinearTrainer.java b/Common/LibLinear/src/main/java/org/tribuo/common/liblinear/LibLinearTrainer.java
index 4f3ecfe5c..84d4d3f66 100644
--- a/Common/LibLinear/src/main/java/org/tribuo/common/liblinear/LibLinearTrainer.java
+++ b/Common/LibLinear/src/main/java/org/tribuo/common/liblinear/LibLinearTrainer.java
@@ -44,6 +44,10 @@
 /**
  * A {@link Trainer} which wraps a liblinear-java trainer.
  * <p>
+ * Note the train method is synchronized on {@code LibLinearTrainer.class} due to a global RNG in liblinear-java.
+ * This is insufficient to ensure reproducibility if liblinear-java is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Fan RE, Chang KW, Hsieh CJ, Wang XR, Lin CJ.
@@ -139,7 +143,10 @@ public LibLinearModel<T> train(Dataset<T> examples, Map<String, Provenance> runP
 
         Pair<FeatureNode[][],double[][]> data = extractData(examples,outputIDInfo,featureIDMap);
 
-        List<de.bwaldvogel.liblinear.Model> models = trainModels(curParams,featureIDMap.size()+1,data.getA(),data.getB());
+        List<de.bwaldvogel.liblinear.Model> models;
+        synchronized (LibLinearTrainer.class) {
+            models = trainModels(curParams, featureIDMap.size() + 1, data.getA(), data.getB());
+        }
 
         return createModel(provenance,featureIDMap,outputIDInfo,models);
     }
diff --git a/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMModel.java b/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMModel.java
index ccb2d418f..ba83b0cea 100644
--- a/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMModel.java
+++ b/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMModel.java
@@ -86,7 +86,7 @@ protected LibSVMModel(String name, ModelProvenance description, ImmutableFeature
     /**
      * Returns an unmodifiable copy of the underlying list of libsvm models.
      * <p>
-     * Deprecated to unify the names across LibLinear, LibSVM and XGBoost.
+     * @deprecated Deprecated to unify the names across LibLinear, LibSVM and XGBoost.
      * @return The underlying model list.
      */
     @Deprecated
diff --git a/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMTrainer.java b/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMTrainer.java
index 48e04ed11..2cfb869bd 100644
--- a/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMTrainer.java
+++ b/Common/LibSVM/src/main/java/org/tribuo/common/libsvm/LibSVMTrainer.java
@@ -36,15 +36,19 @@
 
 import java.time.OffsetDateTime;
 import java.util.ArrayList;
-import java.util.Arrays;
 import java.util.Collections;
 import java.util.List;
 import java.util.Map;
+import java.util.SplittableRandom;
 import java.util.logging.Logger;
 
 /**
  * A trainer that will train using libsvm's Java implementation.
  * <p>
+ * Note the train method is synchronized on {@code LibSVMTrainer.class} due to a global RNG in LibSVM.
+ * This is insufficient to ensure reproducibility if LibSVM is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Chang CC, Lin CJ.
@@ -112,6 +116,11 @@ public abstract class LibSVMTrainer<T extends Output<T>> implements Trainer<T> {
     @Config(description="Generate probability estimates.")
     private boolean probability = false;
 
+    @Config(description="RNG seed.")
+    private long seed = Trainer.DEFAULT_SEED;
+
+    private SplittableRandom rng;
+
     private int trainInvocationCounter = 0;
 
     /**
@@ -123,7 +132,7 @@ protected LibSVMTrainer() {}
      * Constructs a LibSVMTrainer from the parameters.
      * @param parameters The SVM parameters.
      */
-    protected LibSVMTrainer(SVMParameters<T> parameters) {
+    protected LibSVMTrainer(SVMParameters<T> parameters, long seed) {
         this.parameters = parameters.getParameters();
         // Unpack the parameters for the provenance system.
         this.svmType = parameters.getSvmType();
@@ -138,6 +147,8 @@ protected LibSVMTrainer(SVMParameters<T> parameters) {
         this.p = this.parameters.p;
         this.shrinking = this.parameters.shrinking == 1;
         this.probability = this.parameters.probability == 1;
+        this.seed = seed;
+        this.rng = new SplittableRandom(seed);
     }
 
     /**
@@ -158,6 +169,7 @@ public void postConfig() {
         parameters.p = p;
         parameters.shrinking = shrinking ? 1 : 0;
         parameters.probability = probability ? 1 : 0;
+        this.rng = new SplittableRandom(seed);
     }
 
     @Override
@@ -167,6 +179,8 @@ public String toString() {
         buffer.append("LibSVMTrainer(");
         buffer.append("svm_params=");
         buffer.append(SVMParameters.svmParamsToString(parameters));
+        buffer.append(",seed=");
+        buffer.append(seed);
         buffer.append(")");
 
         return buffer.toString();
@@ -185,16 +199,26 @@ public LibSVMModel<T> train(Dataset<T> examples, Map<String, Provenance> runProv
         ImmutableFeatureMap featureIDMap = examples.getFeatureIDMap();
         ImmutableOutputInfo<T> outputIDInfo = examples.getOutputIDInfo();
 
-        TrainerProvenance trainerProvenance = getProvenance();
-        ModelProvenance provenance = new ModelProvenance(LibSVMModel.class.getName(), OffsetDateTime.now(), examples.getProvenance(), trainerProvenance, runProvenance);
-        trainInvocationCounter++;
+        // Creates a new RNG, adds one to the invocation count, generates a local optimiser.
+        TrainerProvenance trainerProvenance;
+        SplittableRandom localRNG;
+        synchronized(this) {
+            localRNG = rng.split();
+            trainerProvenance = getProvenance();
+            trainInvocationCounter++;
+        }
 
         svm_parameter curParams = setupParameters(outputIDInfo);
 
         Pair<svm_node[][],double[][]> data = extractData(examples,outputIDInfo,featureIDMap);
 
-        List<svm_model> models = trainModels(curParams,featureIDMap.size()+1,data.getA(),data.getB());
+        List<svm_model> models;
+        synchronized (LibSVMTrainer.class) {
+            // localRNG is used to seed LibSVM's global RNG before each call to svm.svm_train
+            models = trainModels(curParams, featureIDMap.size() + 1, data.getA(), data.getB(), localRNG);
+        }
 
+        ModelProvenance provenance = new ModelProvenance(LibSVMModel.class.getName(), OffsetDateTime.now(), examples.getProvenance(), trainerProvenance, runProvenance);
         return createModel(provenance,featureIDMap,outputIDInfo,models);
     }
 
@@ -209,17 +233,18 @@ public LibSVMModel<T> train(Dataset<T> examples, Map<String, Provenance> runProv
     protected abstract LibSVMModel<T> createModel(ModelProvenance provenance, ImmutableFeatureMap featureIDMap, ImmutableOutputInfo<T> outputIDInfo, List<svm_model> models);
 
     /**
-     * Train all the liblinear instances necessary for this dataset.
-     * @param curParams The LibLinear parameters.
+     * Train all the LibSVM instances necessary for this dataset.
+     * @param curParams The LibSVM parameters.
      * @param numFeatures The number of features in this dataset.
      * @param features The features themselves.
      * @param outputs The outputs.
-     * @return A list of liblinear models.
+     * @param localRNG The RNG to use for seeding LibSVM's RNG.
+     * @return A list of LibSVM models.
      */
-    protected abstract List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs);
+    protected abstract List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs, SplittableRandom localRNG);
 
     /**
-     * Extracts the features and {@link Output}s in LibLinear's format.
+     * Extracts the features and {@link Output}s in LibSVM's format.
      * @param data The input data.
      * @param outputInfo The output info.
      * @param featureMap The feature info.
diff --git a/Regression/LibLinear/src/main/java/org/tribuo/regression/liblinear/LibLinearRegressionTrainer.java b/Regression/LibLinear/src/main/java/org/tribuo/regression/liblinear/LibLinearRegressionTrainer.java
index 27efdca64..d2e7f5a1c 100644
--- a/Regression/LibLinear/src/main/java/org/tribuo/regression/liblinear/LibLinearRegressionTrainer.java
+++ b/Regression/LibLinear/src/main/java/org/tribuo/regression/liblinear/LibLinearRegressionTrainer.java
@@ -42,6 +42,10 @@
  * <p>
  * This generates an independent liblinear model for each regression dimension.
  * <p>
+ * Note the train method is synchronized on {@code LibLinearTrainer.class} due to a global RNG in liblinear-java.
+ * This is insufficient to ensure reproducibility if liblinear-java is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Fan RE, Chang KW, Hsieh CJ, Wang XR, Lin CJ.
@@ -106,6 +110,9 @@ protected List<Model> trainModels(Parameter curParams, int numFeatures, FeatureN
             data.n = numFeatures;
             data.bias = 1.0;
 
+            // Note this isn't sufficient for reproducibility as it doesn't cope with concurrency.
+            // Concurrency safety is handled by the global lock on LibLinearTrainer.class in LibLinearTrainer.train.
+            Linear.resetRandom();
             models.add(Linear.train(data, curParams));
         }
 
diff --git a/Regression/LibSVM/src/main/java/org/tribuo/regression/libsvm/LibSVMRegressionTrainer.java b/Regression/LibSVM/src/main/java/org/tribuo/regression/libsvm/LibSVMRegressionTrainer.java
index e33e4a361..68bc34f08 100644
--- a/Regression/LibSVM/src/main/java/org/tribuo/regression/libsvm/LibSVMRegressionTrainer.java
+++ b/Regression/LibSVM/src/main/java/org/tribuo/regression/libsvm/LibSVMRegressionTrainer.java
@@ -22,6 +22,7 @@
 import org.tribuo.Example;
 import org.tribuo.ImmutableFeatureMap;
 import org.tribuo.ImmutableOutputInfo;
+import org.tribuo.Trainer;
 import org.tribuo.common.libsvm.LibSVMModel;
 import org.tribuo.common.libsvm.LibSVMTrainer;
 import org.tribuo.common.libsvm.SVMParameters;
@@ -37,11 +38,16 @@
 import java.util.ArrayList;
 import java.util.Collections;
 import java.util.List;
+import java.util.SplittableRandom;
 import java.util.logging.Logger;
 
 /**
  * A trainer for regression models that uses LibSVM. Trains an independent model for each output dimension.
  * <p>
+ * Note the train method is synchronized on {@code LibSVMTrainer.class} due to a global RNG in LibSVM.
+ * This is insufficient to ensure reproducibility if LibSVM is used directly in the same JVM as Tribuo, but
+ * avoids locking on classes Tribuo does not control.
+ * <p>
  * See:
  * <pre>
  * Chang CC, Lin CJ.
@@ -77,16 +83,26 @@ protected LibSVMRegressionTrainer() {}
      * @param parameters The SVM parameters.
      */
     public LibSVMRegressionTrainer(SVMParameters<Regressor> parameters) {
-        this(parameters, false);
+        this(parameters, false, Trainer.DEFAULT_SEED);
     }
 
     /**
-     * Constructs a LibSVMRegressionTrainer using the supplied parameters.
+     * Constructs a LibSVMRegressionTrainer using the supplied parameters and {@link Trainer#DEFAULT_SEED}.
      * @param parameters The SVM parameters.
      * @param standardize Standardize the regression outputs before training.
      */
     public LibSVMRegressionTrainer(SVMParameters<Regressor> parameters, boolean standardize) {
-        super(parameters);
+        this(parameters,standardize,Trainer.DEFAULT_SEED);
+    }
+
+    /**
+     * Constructs a LibSVMRegressionTrainer using the supplied parameters and seed.
+     * @param parameters The SVM parameters.
+     * @param standardize Standardize the regression outputs before training.
+     * @param seed The RNG seed for LibSVM's internal RNG.
+     */
+    public LibSVMRegressionTrainer(SVMParameters<Regressor> parameters, boolean standardize, long seed) {
+        super(parameters,seed);
         this.standardize = standardize;
     }
 
@@ -121,7 +137,7 @@ protected LibSVMModel<Regressor> createModel(ModelProvenance provenance, Immutab
     }
 
     @Override
-    protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs) {
+    protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures, svm_node[][] features, double[][] outputs, SplittableRandom localRNG) {
         ArrayList<svm_model> models = new ArrayList<>();
 
         for (int i = 0; i < outputs.length; i++) {
@@ -132,13 +148,16 @@ protected List<svm_model> trainModels(svm_parameter curParams, int numFeatures,
             if (curParams.gamma == 0) {
                 curParams.gamma = 1.0 / numFeatures;
             }
+            // This is safe because we synchronize on LibSVMTrainer.class in the train method to
+            // ensure there is no concurrent use of the rng.
+            svm.rand.setSeed(localRNG.nextLong());
             if (standardize) {
                 Pair<Double,Double> meanVar = Util.meanAndVariance(outputs[i]);
                 double mean = meanVar.getA();
                 double variance = meanVar.getB();
                 Util.standardizeInPlace(outputs[i],mean,variance);
                 String checkString = svm.svm_check_parameter(problem, curParams);
-                if(checkString != null) {
+                if (checkString != null) {
                     throw new IllegalArgumentException("Error checking SVM parameters: " + checkString);
                 }
                 svm_model trained = svm.svm_train(problem, curParams);