diff --git a/src/IROperator.cpp b/src/IROperator.cpp
index ffd33122d741..a4007c7a8762 100644
--- a/src/IROperator.cpp
+++ b/src/IROperator.cpp
@@ -129,7 +129,8 @@ bool is_const(const Expr &e, int64_t value) {
     } else if (const FloatImm *i = e.as<FloatImm>()) {
         return i->value == value;
     } else if (const Cast *c = e.as<Cast>()) {
-        return is_const(c->value, value);
+        return c->type.can_represent(value) &&
+               is_const(c->value, value);
     } else if (const Broadcast *b = e.as<Broadcast>()) {
         return is_const(b->value, value);
     } else {
diff --git a/test/correctness/solve.cpp b/test/correctness/solve.cpp
index b356d4a1bfe5..69d2c72051a4 100644
--- a/test/correctness/solve.cpp
+++ b/test/correctness/solve.cpp
@@ -164,6 +164,28 @@ void test_positive_const_multiplier_still_rewritten() {
                 (x != 2) || (Mod::make(seven, three) != 0));
 }
 
+// IROperator::is_const(e, -1) used to ignore signedness in casts. When `c`
+// was a cast from `-1` to uint8, Solve.cpp's `f(x) * c cmp b` rewrite for
+// `c == -1` treated `c` as -1 rather than the correct value (255).
+void test_cast_signedness_respected() {
+    // int32(uint8(-1_i8)) == 255, not -1.
+    Expr fake_negative_one = Cast::make(Int(32), Cast::make(UInt(8), -1));
+
+    // Let N = fake_negative_one.
+    // Broken: y < N * x ~> x < -y    ~> 1 < -100      ~> false
+    // Fixed:  y < N * x ~> x * N > y ~> 1 * 255 > 100 ~> true
+    std::map<std::string, Expr> vars{
+        {"x", Expr(1)},
+        {"y", Expr(100)},
+    };
+    check_solve_equivalent(y < fake_negative_one * x, vars);
+    check_solve_equivalent(y <= fake_negative_one * x, vars);
+    check_solve_equivalent(y > fake_negative_one * x, vars);
+    check_solve_equivalent(y >= fake_negative_one * x, vars);
+    check_solve_equivalent(y == fake_negative_one * x, vars);
+    check_solve_equivalent(y != fake_negative_one * x, vars);
+}
+
 // Solver used to rewrite `f(x) + f(x) -> f(x) * 2` via `operator*(Expr, int)`,
 // which rejects constants that don't fit in the expression type. For UInt(1),
 // the literal 2 isn't representable, aborting the whole solve. Use Mul::make
@@ -610,6 +632,7 @@ int main(int argc, char **argv) {
     test_unsigned_equality_still_rearranged();
     test_nonconstant_multiplier_not_rewritten();
     test_positive_const_multiplier_still_rewritten();
+    test_cast_signedness_respected();
     test_solve_does_not_abort_on_narrow_self_add();
     test_narrow_div_add_equivalence();
     test_simplify_preserves_float_to_uint_cast_chain();