diff --git a/internal/literals/literals.go b/internal/literals/literals.go
index 9695adc..8619ad9 100644
--- a/internal/literals/literals.go
+++ b/internal/literals/literals.go
@@ -28,13 +28,8 @@ import (
 // should largely stop being used.
 const maxSizeBytes = 2 << 10 // KiB
 
-func randObfuscator() obfuscator {
-	randPos := mathrand.Intn(len(obfuscators))
-	return obfuscators[randPos]
-}
-
 // Obfuscate replaces literals with obfuscated anonymous functions.
-func Obfuscate(file *ast.File, info *types.Info, fset *token.FileSet, linkStrings map[*types.Var]string) *ast.File {
+func Obfuscate(obfRand *mathrand.Rand, file *ast.File, info *types.Info, linkStrings map[*types.Var]string) *ast.File {
 	pre := func(cursor *astutil.Cursor) bool {
 		switch node := cursor.Node().(type) {
 		case *ast.GenDecl:
@@ -72,7 +67,7 @@ func Obfuscate(file *ast.File, info *types.Info, fset *token.FileSet, linkString
 				return true
 			}
 
-			cursor.Replace(withPos(obfuscateString(value), node.Pos()))
+			cursor.Replace(withPos(obfuscateString(obfRand, value), node.Pos()))
 
 			return true
 		}
@@ -89,7 +84,7 @@ func Obfuscate(file *ast.File, info *types.Info, fset *token.FileSet, linkString
 			}
 
 			if child, ok := node.X.(*ast.CompositeLit); ok {
-				newnode := handleCompositeLiteral(true, child, info)
+				newnode := handleCompositeLiteral(obfRand, true, child, info)
 				if newnode != nil {
 					cursor.Replace(newnode)
 				}
@@ -108,7 +103,7 @@ func Obfuscate(file *ast.File, info *types.Info, fset *token.FileSet, linkString
 				return true
 			}
 
-			newnode := handleCompositeLiteral(false, node, info)
+			newnode := handleCompositeLiteral(obfRand, false, node, info)
 			if newnode != nil {
 				cursor.Replace(newnode)
 			}
@@ -126,7 +121,7 @@ func Obfuscate(file *ast.File, info *types.Info, fset *token.FileSet, linkString
 //
 // If the input is not a byte slice or array, the node is returned as-is and
 // the second return value will be false.
-func handleCompositeLiteral(isPointer bool, node *ast.CompositeLit, info *types.Info) ast.Node {
+func handleCompositeLiteral(obfRand *mathrand.Rand, isPointer bool, node *ast.CompositeLit, info *types.Info) ast.Node {
 	if len(node.Elts) == 0 || len(node.Elts) > maxSizeBytes {
 		return nil
 	}
@@ -169,10 +164,10 @@ func handleCompositeLiteral(isPointer bool, node *ast.CompositeLit, info *types.
 	}
 
 	if arrayLen > 0 {
-		return withPos(obfuscateByteArray(isPointer, data, arrayLen), node.Pos())
+		return withPos(obfuscateByteArray(obfRand, isPointer, data, arrayLen), node.Pos())
 	}
 
-	return withPos(obfuscateByteSlice(isPointer, data), node.Pos())
+	return withPos(obfuscateByteSlice(obfRand, isPointer, data), node.Pos())
 }
 
 // withPos sets any token.Pos fields under node which affect printing to pos.
@@ -220,18 +215,18 @@ func withPos(node ast.Node, pos token.Pos) ast.Node {
 	return node
 }
 
-func obfuscateString(data string) *ast.CallExpr {
-	obfuscator := randObfuscator()
-	block := obfuscator.obfuscate([]byte(data))
+func obfuscateString(obfRand *mathrand.Rand, data string) *ast.CallExpr {
+	obfuscator := obfuscators[obfRand.Intn(len(obfuscators))]
+	block := obfuscator.obfuscate(obfRand, []byte(data))
 
 	block.List = append(block.List, ah.ReturnStmt(ah.CallExpr(ast.NewIdent("string"), ast.NewIdent("data"))))
 
 	return ah.LambdaCall(ast.NewIdent("string"), block)
 }
 
-func obfuscateByteSlice(isPointer bool, data []byte) *ast.CallExpr {
-	obfuscator := randObfuscator()
-	block := obfuscator.obfuscate(data)
+func obfuscateByteSlice(obfRand *mathrand.Rand, isPointer bool, data []byte) *ast.CallExpr {
+	obfuscator := obfuscators[obfRand.Intn(len(obfuscators))]
+	block := obfuscator.obfuscate(obfRand, data)
 
 	if isPointer {
 		block.List = append(block.List, ah.ReturnStmt(&ast.UnaryExpr{
@@ -247,9 +242,9 @@ func obfuscateByteSlice(isPointer bool, data []byte) *ast.CallExpr {
 	return ah.LambdaCall(&ast.ArrayType{Elt: ast.NewIdent("byte")}, block)
 }
 
-func obfuscateByteArray(isPointer bool, data []byte, length int64) *ast.CallExpr {
-	obfuscator := randObfuscator()
-	block := obfuscator.obfuscate(data)
+func obfuscateByteArray(obfRand *mathrand.Rand, isPointer bool, data []byte, length int64) *ast.CallExpr {
+	obfuscator := obfuscators[obfRand.Intn(len(obfuscators))]
+	block := obfuscator.obfuscate(obfRand, data)
 
 	arrayType := &ast.ArrayType{
 		Len: ah.IntLit(int(length)),
diff --git a/internal/literals/obfuscators.go b/internal/literals/obfuscators.go
index 3f13f11..24e66a6 100644
--- a/internal/literals/obfuscators.go
+++ b/internal/literals/obfuscators.go
@@ -12,7 +12,7 @@ import (
 
 // obfuscator takes a byte slice and converts it to a ast.BlockStmt
 type obfuscator interface {
-	obfuscate(data []byte) *ast.BlockStmt
+	obfuscate(obfRand *mathrand.Rand, data []byte) *ast.BlockStmt
 }
 
 // obfuscators contains all types which implement the obfuscator Interface
@@ -24,33 +24,17 @@ var obfuscators = []obfuscator{
 	// seed{}, TODO: re-enable once https://go.dev/issue/47631 is fixed in Go 1.20
 }
 
-// If math/rand.Seed() is not called, the generator behaves as if seeded by rand.Seed(1),
-// so the generator is deterministic.
-
-// genRandBytes return a random []byte with the length of size.
-func genRandBytes(buffer []byte) {
-	if _, err := mathrand.Read(buffer); err != nil {
-		panic(fmt.Sprintf("couldn't generate random key:  %v", err))
-	}
-}
-
-func genRandByte() byte {
-	bytes := make([]byte, 1)
-	genRandBytes(bytes)
-	return bytes[0]
-}
-
-func genRandIntSlice(max, count int) []int {
+func genRandIntSlice(obfRand *mathrand.Rand, max, count int) []int {
 	indexes := make([]int, count)
 	for i := 0; i < count; i++ {
-		indexes[i] = mathrand.Intn(max)
+		indexes[i] = obfRand.Intn(max)
 	}
 	return indexes
 }
 
-func randOperator() token.Token {
+func randOperator(obfRand *mathrand.Rand) token.Token {
 	operatorTokens := [...]token.Token{token.XOR, token.ADD, token.SUB}
-	return operatorTokens[mathrand.Intn(len(operatorTokens))]
+	return operatorTokens[obfRand.Intn(len(operatorTokens))]
 }
 
 func evalOperator(t token.Token, x, y byte) byte {
diff --git a/internal/literals/seed.go b/internal/literals/seed.go
index 32a2d01..d842702 100644
--- a/internal/literals/seed.go
+++ b/internal/literals/seed.go
@@ -6,6 +6,7 @@ package literals
 import (
 	"go/ast"
 	"go/token"
+	mathrand "math/rand"
 
 	ah "mvdan.cc/garble/internal/asthelper"
 )
@@ -15,11 +16,11 @@ type seed struct{}
 // check that the obfuscator interface is implemented
 var _ obfuscator = seed{}
 
-func (seed) obfuscate(data []byte) *ast.BlockStmt {
-	seed := genRandByte()
+func (seed) obfuscate(obfRand *mathrand.Rand, data []byte) *ast.BlockStmt {
+	seed := byte(obfRand.Uint32())
 	originalSeed := seed
 
-	op := randOperator()
+	op := randOperator(obfRand)
 
 	var callExpr *ast.CallExpr
 	for i, b := range data {
diff --git a/internal/literals/shuffle.go b/internal/literals/shuffle.go
index f9d3168..be1de5a 100644
--- a/internal/literals/shuffle.go
+++ b/internal/literals/shuffle.go
@@ -16,21 +16,21 @@ type shuffle struct{}
 // check that the obfuscator interface is implemented
 var _ obfuscator = shuffle{}
 
-func (shuffle) obfuscate(data []byte) *ast.BlockStmt {
+func (shuffle) obfuscate(obfRand *mathrand.Rand, data []byte) *ast.BlockStmt {
 	key := make([]byte, len(data))
-	genRandBytes(key)
+	obfRand.Read(key)
 
 	fullData := make([]byte, len(data)+len(key))
 	operators := make([]token.Token, len(fullData))
 	for i := range operators {
-		operators[i] = randOperator()
+		operators[i] = randOperator(obfRand)
 	}
 
 	for i, b := range key {
 		fullData[i], fullData[i+len(data)] = evalOperator(operators[i], data[i], b), b
 	}
 
-	shuffledIdxs := mathrand.Perm(len(fullData))
+	shuffledIdxs := obfRand.Perm(len(fullData))
 
 	shuffledFullData := make([]byte, len(fullData))
 	for i, b := range fullData {
diff --git a/internal/literals/simple.go b/internal/literals/simple.go
index 1a776aa..ee2c0e5 100644
--- a/internal/literals/simple.go
+++ b/internal/literals/simple.go
@@ -6,6 +6,7 @@ package literals
 import (
 	"go/ast"
 	"go/token"
+	mathrand "math/rand"
 
 	ah "mvdan.cc/garble/internal/asthelper"
 )
@@ -15,11 +16,11 @@ type simple struct{}
 // check that the obfuscator interface is implemented
 var _ obfuscator = simple{}
 
-func (simple) obfuscate(data []byte) *ast.BlockStmt {
+func (simple) obfuscate(obfRand *mathrand.Rand, data []byte) *ast.BlockStmt {
 	key := make([]byte, len(data))
-	genRandBytes(key)
+	obfRand.Read(key)
 
-	op := randOperator()
+	op := randOperator(obfRand)
 	for i, b := range key {
 		data[i] = evalOperator(op, data[i], b)
 	}
diff --git a/internal/literals/split.go b/internal/literals/split.go
index 314ca22..584c5c2 100644
--- a/internal/literals/split.go
+++ b/internal/literals/split.go
@@ -23,14 +23,14 @@ type split struct{}
 // check that the obfuscator interface is implemented
 var _ obfuscator = split{}
 
-func splitIntoRandomChunks(data []byte) [][]byte {
+func splitIntoRandomChunks(obfRand *mathrand.Rand, data []byte) [][]byte {
 	if len(data) == 1 {
 		return [][]byte{data}
 	}
 
 	var chunks [][]byte
 	for len(data) > 0 {
-		chunkSize := 1 + mathrand.Intn(maxChunkSize)
+		chunkSize := 1 + obfRand.Intn(maxChunkSize)
 		if chunkSize > len(data) {
 			chunkSize = len(data)
 		}
@@ -51,8 +51,8 @@ func splitIntoOneByteChunks(data []byte) [][]byte {
 
 // Shuffles the passed array and returns it back.
 // Applies for inline declaration of randomly shuffled statement arrays
-func shuffleStmts(stmts ...ast.Stmt) []ast.Stmt {
-	mathrand.Shuffle(len(stmts), func(i, j int) {
+func shuffleStmts(obfRand *mathrand.Rand, stmts ...ast.Stmt) []ast.Stmt {
+	obfRand.Shuffle(len(stmts), func(i, j int) {
 		stmts[i], stmts[j] = stmts[j], stmts[i]
 	})
 	return stmts
@@ -69,33 +69,33 @@ func encryptChunks(chunks [][]byte, op token.Token, key byte) {
 	}
 }
 
-func (split) obfuscate(data []byte) *ast.BlockStmt {
+func (split) obfuscate(obfRand *mathrand.Rand, data []byte) *ast.BlockStmt {
 	var chunks [][]byte
 	// Short arrays should be divided into single-byte fragments
 	if len(data)/maxChunkSize < minCaseCount {
 		chunks = splitIntoOneByteChunks(data)
 	} else {
-		chunks = splitIntoRandomChunks(data)
+		chunks = splitIntoRandomChunks(obfRand, data)
 	}
 
 	// Generate indexes for cases chunk count + 1 decrypt case + 1 exit case
-	indexes := mathrand.Perm(len(chunks) + 2)
+	indexes := obfRand.Perm(len(chunks) + 2)
 
-	decryptKeyInitial := genRandByte()
+	decryptKeyInitial := byte(obfRand.Uint32())
 	decryptKey := decryptKeyInitial
 	// Calculate decrypt key based on indexes and position. Ignore exit index
 	for i, index := range indexes[:len(indexes)-1] {
 		decryptKey ^= byte(index * i)
 	}
 
-	op := randOperator()
+	op := randOperator(obfRand)
 	encryptChunks(chunks, op, decryptKey)
 
 	decryptIndex := indexes[len(indexes)-2]
 	exitIndex := indexes[len(indexes)-1]
 	switchCases := []ast.Stmt{&ast.CaseClause{
 		List: []ast.Expr{ah.IntLit(decryptIndex)},
-		Body: shuffleStmts(
+		Body: shuffleStmts(obfRand,
 			&ast.AssignStmt{
 				Lhs: []ast.Expr{ast.NewIdent("i")},
 				Tok: token.ASSIGN,
@@ -142,7 +142,7 @@ func (split) obfuscate(data []byte) *ast.BlockStmt {
 
 		switchCases = append(switchCases, &ast.CaseClause{
 			List: []ast.Expr{ah.IntLit(index)},
-			Body: shuffleStmts(
+			Body: shuffleStmts(obfRand,
 				&ast.AssignStmt{
 					Lhs: []ast.Expr{ast.NewIdent("i")},
 					Tok: token.ASSIGN,
@@ -206,7 +206,7 @@ func (split) obfuscate(data []byte) *ast.BlockStmt {
 				},
 				&ast.SwitchStmt{
 					Tag:  ast.NewIdent("i"),
-					Body: ah.BlockStmt(shuffleStmts(switchCases...)...),
+					Body: ah.BlockStmt(shuffleStmts(obfRand, switchCases...)...),
 				}),
 		},
 	)
diff --git a/internal/literals/swap.go b/internal/literals/swap.go
index 5fd88c1..b5c50fd 100644
--- a/internal/literals/swap.go
+++ b/internal/literals/swap.go
@@ -45,12 +45,12 @@ func positionsToSlice(data []int) *ast.CompositeLit {
 }
 
 // Generates a random even swap count based on the length of data
-func generateSwapCount(dataLen int) int {
+func generateSwapCount(obfRand *mathrand.Rand, dataLen int) int {
 	swapCount := dataLen
 
 	maxExtraPositions := dataLen / 2 // Limit the number of extra positions to half the data length
 	if maxExtraPositions > 1 {
-		swapCount += mathrand.Intn(maxExtraPositions)
+		swapCount += obfRand.Intn(maxExtraPositions)
 	}
 	if swapCount%2 != 0 { // Swap count must be even
 		swapCount++
@@ -58,13 +58,13 @@ func generateSwapCount(dataLen int) int {
 	return swapCount
 }
 
-func (swap) obfuscate(data []byte) *ast.BlockStmt {
-	swapCount := generateSwapCount(len(data))
-	shiftKey := genRandByte()
+func (swap) obfuscate(obfRand *mathrand.Rand, data []byte) *ast.BlockStmt {
+	swapCount := generateSwapCount(obfRand, len(data))
+	shiftKey := byte(obfRand.Uint32())
 
-	op := randOperator()
+	op := randOperator(obfRand)
 
-	positions := genRandIntSlice(len(data), swapCount)
+	positions := genRandIntSlice(obfRand, len(data), swapCount)
 	for i := len(positions) - 2; i >= 0; i -= 2 {
 		// Generate local key for xor based on random key and byte position
 		localKey := byte(i) + byte(positions[i]^positions[i+1]) + shiftKey
diff --git a/main.go b/main.go
index 88befa0..8197c32 100644
--- a/main.go
+++ b/main.go
@@ -5,7 +5,7 @@ package main
 
 import (
 	"bytes"
-	"crypto/rand"
+	cryptorand "crypto/rand"
 	"encoding/base64"
 	"encoding/binary"
 	"encoding/gob"
@@ -78,7 +78,7 @@ func (f seedFlag) String() string {
 func (f *seedFlag) Set(s string) error {
 	if s == "random" {
 		f.bytes = make([]byte, 16) // random 128 bit seed
-		if _, err := rand.Read(f.bytes); err != nil {
+		if _, err := cryptorand.Read(f.bytes); err != nil {
 			return fmt.Errorf("error generating random seed: %v", err)
 		}
 	} else {
@@ -90,6 +90,8 @@ func (f *seedFlag) Set(s string) error {
 			return fmt.Errorf("error decoding seed: %v", err)
 		}
 
+		// TODO: Note that we always use 8 bytes; any bytes after that are
+		// entirely ignored. That may be confusing to the end user.
 		if len(seed) < 8 {
 			return fmt.Errorf("-seed needs at least 8 bytes, have %d", len(seed))
 		}
@@ -151,6 +153,13 @@ var (
 
 	// Basic information about the package being currently compiled or linked.
 	curPkg *listedPackage
+
+	// obfRand is initialized by transformCompile and used during obfuscation.
+	// It is left nil at init time, so that we only use it after it has been
+	// properly initialized with a deterministic seed.
+	// It must only be used for deterministic obfuscation;
+	// if it is used for any other purpose, we may lose determinism.
+	obfRand *mathrand.Rand
 )
 
 type importerWithMap func(path, dir string, mode types.ImportMode) (*types.Package, error)
@@ -848,7 +857,7 @@ func transformCompile(args []string) ([]string, error) {
 		randSeed = flagSeed.bytes
 	}
 	// log.Printf("seeding math/rand with %x\n", randSeed)
-	mathrand.Seed(int64(binary.BigEndian.Uint64(randSeed)))
+	obfRand = mathrand.New(mathrand.NewSource(int64(binary.BigEndian.Uint64(randSeed))))
 
 	if err := tf.prefillObjectMaps(files); err != nil {
 		return nil, err
@@ -1652,7 +1661,7 @@ func (tf *transformer) transformGo(file *ast.File) *ast.File {
 	// because obfuscated literals sometimes escape to heap,
 	// and that's not allowed in the runtime itself.
 	if flagLiterals && curPkg.ToObfuscate {
-		file = literals.Obfuscate(file, tf.info, fset, tf.linkerVariableStrings)
+		file = literals.Obfuscate(obfRand, file, tf.info, tf.linkerVariableStrings)
 
 		// some imported constants might not be needed anymore, remove unnecessary imports
 		tf.removeUnnecessaryImports(file)
diff --git a/main_test.go b/main_test.go
index 0a001ed..f5825b2 100644
--- a/main_test.go
+++ b/main_test.go
@@ -230,9 +230,11 @@ func bincmp(ts *testscript.TestScript, neg bool, args []string) {
 	}
 }
 
+var testRand = mathrand.New(mathrand.NewSource(time.Now().UnixNano()))
+
 func generateStringLit(size int) *ast.BasicLit {
 	buffer := make([]byte, size)
-	_, err := mathrand.Read(buffer)
+	_, err := testRand.Read(buffer)
 	if err != nil {
 		panic(err)
 	}
@@ -253,7 +255,7 @@ func generateLiterals(ts *testscript.TestScript, neg bool, args []string) {
 	// Add 100 randomly small literals.
 	var statements []ast.Stmt
 	for i := 0; i < 100; i++ {
-		literal := generateStringLit(1 + mathrand.Intn(255))
+		literal := generateStringLit(1 + testRand.Intn(255))
 		statements = append(statements, &ast.AssignStmt{
 			Lhs: []ast.Expr{ast.NewIdent("_")},
 			Tok: token.ASSIGN,