diff --git a/source/air/src/def.rs b/source/air/src/def.rs
index fac4314b1c..702e0d82b9 100644
--- a/source/air/src/def.rs
+++ b/source/air/src/def.rs
@@ -12,6 +12,7 @@ pub const CHOOSE: &str = "%%choose%%";
 pub const HOLE: &str = "%%hole%%";
 pub const APPLY: &str = "%%apply%%";
 pub const TEMP: &str = "%%x%%";
+pub const MANGLED_SYMBOL_PREFIX: &str = "%%verus_mangled%%";
 pub const SKOLEM_ID_PREFIX: &str = "skolem";
 pub const ARRAY_QID: &str = "__AIR_ARRAY_QID__";
 
diff --git a/source/air/src/parser.rs b/source/air/src/parser.rs
index 671d52c91a..34af6e9815 100644
--- a/source/air/src/parser.rs
+++ b/source/air/src/parser.rs
@@ -11,13 +11,9 @@ use sise::Node;
 use std::io::Write;
 use std::sync::Arc;
 
-// Following SMT-LIB syntax specification
-fn is_symbol_char(c: char) -> bool {
-    c.is_ascii_alphanumeric() || "~!@$%^&*_-+=<>.?/".contains(c)
-}
-
+// AIR symbols may contain Unicode or other non-SMT characters; the printer will mangle them before sending to the solver.
 fn is_symbol(s: &String) -> bool {
-    s.len() > 0 && s.chars().all(is_symbol_char)
+    s.len() > 0
 }
 
 fn is_bitvec(nodes: &Vec<Node>) -> Option<u32> {
diff --git a/source/air/src/printer.rs b/source/air/src/printer.rs
index cc88fc06ca..fa18d9393f 100644
--- a/source/air/src/printer.rs
+++ b/source/air/src/printer.rs
@@ -4,11 +4,78 @@ use crate::ast::{
     Typs, UnaryOp,
 };
 use crate::context::SmtSolver;
-use crate::def::mk_skolem_id;
+use crate::def::{MANGLED_SYMBOL_PREFIX, mk_skolem_id};
 use crate::util::vec_map;
 use sise::{Node, Writer};
 use std::sync::Arc;
 
+fn is_smt_simple_symbol_char(c: char) -> bool {
+    c.is_ascii_alphanumeric()
+        || matches!(
+            c,
+            '~' | '!'
+                | '@'
+                | '$'
+                | '%'
+                | '^'
+                | '&'
+                | '*'
+                | '_'
+                | '-'
+                | '+'
+                | '='
+                | '<'
+                | '>'
+                | '.'
+                | '?'
+                | '/'
+        )
+}
+
+fn is_valid_smt_atom(s: &str) -> bool {
+    if let Some(rest) = s.strip_prefix(':') {
+        if !rest.is_empty() && rest.chars().all(is_smt_simple_symbol_char) {
+            return true;
+        }
+    }
+    if s.bytes().all(|b| b.is_ascii_digit()) {
+        return true;
+    }
+    if let Some((lhs, rhs)) = s.split_once('.') {
+        if !lhs.is_empty()
+            && !rhs.is_empty()
+            && lhs.bytes().all(|b| b.is_ascii_digit())
+            && rhs.bytes().all(|b| b.is_ascii_digit())
+        {
+            return true;
+        }
+    }
+    let Some(first) = s.chars().next() else {
+        return false;
+    };
+    !first.is_ascii_digit()
+        && is_smt_simple_symbol_char(first)
+        && s.chars().skip(1).all(is_smt_simple_symbol_char)
+}
+
+fn mangle_smt_atom(s: &str) -> String {
+    use std::fmt::Write as _;
+    let mut out = String::from(MANGLED_SYMBOL_PREFIX);
+    if s.is_empty() {
+        out.push('_');
+        return out;
+    }
+    for ch in s.chars() {
+        write!(&mut out, "_{:x}", ch as u32).expect("Writing to String should not fail");
+    }
+    out
+}
+
+pub fn sanitize_ident(s: &str) -> Ident {
+    Arc::new(if is_valid_smt_atom(s) { s.to_string() } else { mangle_smt_atom(s) })
+}
+
+#[inline]
 pub fn str_to_node(s: &str) -> Node {
     Node::Atom(s.to_string())
 }
@@ -589,7 +656,9 @@ impl NodeWriter {
             sise::SpacedStringWriterNodeOptions { break_line_len: if brk { 0 } else { break_len } };
         match node {
             Node::Atom(a) => {
-                writer.write_atom(a, opts).unwrap();
+                let a = if is_valid_smt_atom(a) { a.clone() } else { mangle_smt_atom(a) };
+                assert!(sise::check_atom(&a), "invalid atom: {}", a);
+                writer.write_atom(&a, opts).unwrap();
             }
             Node::List(l) => {
                 writer.begin_list(opts).unwrap();
diff --git a/source/air/src/tests.rs b/source/air/src/tests.rs
index b94cfacaff..99ee3cf5e8 100644
--- a/source/air/src/tests.rs
+++ b/source/air/src/tests.rs
@@ -4,10 +4,20 @@ use crate::messages::Reporter;
 #[allow(unused_imports)]
 use crate::parser::Parser;
 #[allow(unused_imports)]
-use crate::printer::{macro_push_node, str_to_node};
+use crate::printer::{NodeWriter, macro_push_node, str_to_node};
 #[allow(unused_imports)]
 use sise::Node;
 
+#[test]
+fn node_writer_mangles_invalid_atoms() {
+    for atom in ["r#for$", "r#in$", "r#bad"] {
+        let mut nw = NodeWriter::new();
+        let s = nw.node_to_string_indent(&String::new(), &Node::Atom(atom.to_string()));
+        assert!(s.contains(crate::def::MANGLED_SYMBOL_PREFIX));
+        assert!(!s.contains('#'));
+    }
+}
+
 #[allow(dead_code)]
 fn run_nodes_as_test(should_typecheck: bool, should_be_valid: bool, nodes: &[Node]) {
     let message_interface = std::sync::Arc::new(crate::messages::AirMessageInterface {});
diff --git a/source/rust_verify_test/tests/regression.rs b/source/rust_verify_test/tests/regression.rs
index 5dc17781ad..965aeffb1e 100644
--- a/source/rust_verify_test/tests/regression.rs
+++ b/source/rust_verify_test/tests/regression.rs
@@ -1,4 +1,5 @@
 #![feature(rustc_private)]
+#![allow(mixed_script_confusables)]
 #[macro_use]
 mod common;
 use common::*;
@@ -1587,3 +1588,132 @@ test_verify_one_file! {
         }
     } => Ok(())
 }
+
+test_verify_one_file! {
+    #[test] raw_identifiers_in_smt_2221 verus_code! {
+        // Raw identifiers (r#keyword) must be sanitized before reaching SMT-LIB,
+        // since '#' is not a valid SMT-LIB simple-symbol character.
+        // Test a representative set of Rust strict keywords as function names,
+        // parameters, local variables, struct fields, and enum variants.
+
+        struct r#struct {
+            r#type: u32,
+            r#loop: u32,
+        }
+
+        enum r#enum {
+            r#match,
+            r#impl,
+        }
+
+        fn r#match(
+            r#in: u32,
+            r#out: u32,
+            r#for: u32,
+            r#while: u32,
+            r#return: u32,
+            r#break: u32,
+            r#continue: u32,
+            r#if: u32,
+            r#else: u32,
+            r#let: u32,
+            r#mut: u32,
+            r#ref: u32,
+            r#move: u32,
+            r#const: u32,
+            r#static: u32,
+            r#unsafe: u32,
+            r#where: u32,
+            r#async: u32,
+            r#await: u32,
+            r#dyn: u32,
+            r#as: u32,
+            r#use: u32,
+            r#pub: u32,
+            r#mod: u32,
+        ) -> u32
+            requires r#in == r#out
+        {
+            let r#type = r#in;
+            let r#loop = r#for;
+            let r#match = r#while;
+            let r#impl = r#return;
+            let r#enum = r#break;
+            let r#struct = r#continue;
+            let r#trait = r#if;
+            let r#let = r#else;
+            let r#mut = r#let;
+            let r#ref = r#mut;
+            let r#move = r#ref;
+            let r#const = r#move;
+            let r#static = r#const;
+            let r#unsafe = r#static;
+            let r#where = r#unsafe;
+            let r#async = r#where;
+            let r#await = r#async;
+            let r#dyn = r#await;
+            let r#as = r#dyn;
+            let r#use = r#as;
+            let r#pub = r#use;
+            let r#mod = r#pub;
+            assert(r#type == r#out);
+            r#type
+        }
+
+        fn raw_param(r#for: u32) -> u32 { r#for }
+    } => Ok(())
+}
+
+test_verify_one_file! {
+    #[test] unicode_identifiers_in_smt_2221 verus_code! {
+        // Unicode identifiers must be sanitized before reaching AIR/SMT-LIB,
+        // since non-ASCII characters are not valid in SMT-LIB simple symbols.
+        // Test unicode in types, fields, enums, traits, generics, functions,
+        // parameters, locals, constants, and spec/proof functions.
+
+        struct Wrapperα<Tβ> {
+            payloadγ: Tβ,
+            extraδ: u32,
+        }
+
+        enum Choiceε<Tζ> {
+            OptionAη,
+            OptionBθ(Tζ),
+        }
+
+        trait Traitι {
+            spec fn methodκ(&self) -> int;
+        }
+
+        impl Traitι for Wrapperα<u32> {
+            spec fn methodκ(&self) -> int { self.payloadγ as int + self.extraδ as int }
+        }
+
+        spec fn αβγ(x: u32) -> u32 { x }
+
+        const MAXλ: u32 = 42;
+
+        proof fn test_unicode<Genμ>(item: Wrapperα<u32>, choice: Choiceε<u32>)
+            requires item.payloadγ == MAXλ
+        {
+            let baseν = item.payloadγ;
+            let scaledξ = αβγ(1u32);
+            let valueο = MAXλ;
+            assert(baseν == MAXλ);
+            assert(scaledξ == 1u32);
+            assert(valueο == 42);
+            assert(item.methodκ() == MAXλ as int + item.extraδ as int);
+        }
+    } => Ok(_err) => {
+        // Verification succeeds; the only warnings are the mixed_script_confusables
+        // lint about using non-ASCII identifiers, which is a Rust lint unrelated to AIR sanitization.
+        assert!(_err.errors.is_empty(), "unexpected errors: {:?}", _err.errors);
+        for w in &_err.warnings {
+            let msg = &w.message;
+            assert!(
+                msg.contains("mixed script confusables") || msg.contains("warning emitted"),
+                "unexpected warning: {:?}", msg
+            );
+        }
+    }
+}
diff --git a/source/vir/src/def.rs b/source/vir/src/def.rs
index 2d26041d01..fc281774ad 100644
--- a/source/vir/src/def.rs
+++ b/source/vir/src/def.rs
@@ -681,7 +681,12 @@ pub fn encode_dt_as_path(dt: &Dt) -> Path {
 
 pub fn variant_ident(dt: &Dt, variant: &str) -> Ident {
     let path = encode_dt_as_path(dt);
-    Arc::new(format!("{}{}{}", path_to_string(&path), VARIANT_SEPARATOR, variant))
+    air::printer::sanitize_ident(&format!(
+        "{}{}{}",
+        path_to_string(&path),
+        VARIANT_SEPARATOR,
+        variant
+    ))
 }
 
 pub fn is_variant_ident(datatype: &Dt, variant: &str) -> Ident {
@@ -694,7 +699,7 @@ pub fn variant_field_ident_internal(
     field: &Ident,
     internal: bool,
 ) -> Ident {
-    Arc::new(format!(
+    air::printer::sanitize_ident(&format!(
         "{}{}{}{}{}",
         path_to_string(path),
         VARIANT_SEPARATOR,