FIX: preserve unknown characters in BrailleConverter (#2309)

Signed-off-by: fei <204683769+feiiiiii5@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 46d2071a-7f67-4855-84da-d0f19efb8a53

Author: feiiiiii5

pyrit/converter/braille_converter.py

@@ -11,12 +11,17 @@ class BrailleConverter(Converter):
     Converts text into Braille Unicode representation.
 
     This converter transforms standard text into Braille patterns using Unicode
-    Braille characters (U+2800 to U+28FF). It supports lowercase and uppercase
-    letters, numbers, common punctuation, and spaces. Uppercase letters are
-    prefixed with the Braille capitalization indicator.
-
-    The Braille mapping is based on the implementation from Garak:
+    Braille characters (U+2800 to U+28FF). Every printable ASCII character is
+    mapped: letters, digits, punctuation, and symbols. Uppercase letters are
+    prefixed with the Braille capitalization indicator and digit runs with the
+    number indicator. Characters with no Braille mapping (accented and CJK
+    letters, emoji, control characters) pass through unchanged rather than being
+    dropped, so the encoded prompt keeps its meaning.
+
+    The letter, digit, and core punctuation mappings are based on the
+    implementation from Garak:
     https://github.com/NVIDIA/garak/blob/main/garak/probes/encoding.py
+    The ASCII symbol cells follow Unified English Braille (UEB).
 
     Note: This converter is useful for testing how AI systems handle Braille-encoded
     text, which can be used to obfuscate potentially harmful content.
@@ -99,6 +104,27 @@ def _get_braile(self, text: str) -> str:
             ";": "\u2806",
             "(": "\u2836",
             ")": "\u2836",
+            # Remaining printable ASCII, as two-cell UEB symbol sequences.
+            "@": "\u2808\u2801",
+            "#": "\u2838\u2839",
+            "%": "\u2828\u2834",
+            "&": "\u2808\u282f",
+            "*": "\u2810\u2814",
+            "+": "\u2810\u2816",
+            "<": "\u2808\u2823",
+            "=": "\u2810\u2836",
+            ">": "\u2808\u281c",
+            '"': "\u2820\u2836",
+            "[": "\u2828\u2823",
+            "]": "\u2828\u281c",
+            "\\": "\u2838\u2821",
+            "^": "\u2808\u2822",
+            "_": "\u2828\u2824",
+            "`": "\u2828\u2821",
+            "{": "\u2838\u2823",
+            "}": "\u2838\u281c",
+            "|": "\u2838\u2833",
+            "~": "\u2808\u2814",
             "1": "\u2801",
             "2": "\u2803",
             "3": "\u2809",
@@ -112,23 +138,21 @@ def _get_braile(self, text: str) -> str:
             " ": " ",
         }
         number_punctuations = [".", ",", "-", "/", "$"]
-        escape_characters = ["\n", "\r", "\t"]
 
         output = ""
 
         is_number = False
         for char in text:
-            if char in escape_characters:
-                output += char
-            elif char.isupper():
-                if char.lower() in character_unicodes:
-                    output += character_unicodes["caps"]
-                    output += character_unicodes[char.lower()]
-            elif char in character_unicodes:
+            if char in character_unicodes:
                 if char.isdigit() and not is_number:
                     is_number = True
                     output += character_unicodes["num"]
                 output += character_unicodes[char]
+            elif char.isupper() and char.lower() in character_unicodes:
+                output += character_unicodes["caps"]
+                output += character_unicodes[char.lower()]
+            else:
+                output += char
             if is_number and not char.isdigit() and char not in number_punctuations:
                 is_number = False
 

tests/unit/converter/test_braille_converter.py

@@ -5,6 +5,30 @@
 
 from pyrit.converter import BrailleConverter, ConverterResult
 
+# Printable ASCII symbols and their two-cell Unified English Braille sequences.
+UEB_SYMBOL_CELLS = {
+    "@": "\u2808\u2801",  # dot 4, dot 1
+    "#": "\u2838\u2839",  # dots 456, dots 1456
+    "%": "\u2828\u2834",  # dots 46, dots 356
+    "&": "\u2808\u282f",  # dot 4, dots 12346
+    "*": "\u2810\u2814",  # dot 5, dots 35
+    "+": "\u2810\u2816",  # dot 5, dots 235
+    "<": "\u2808\u2823",  # dot 4, dots 126
+    "=": "\u2810\u2836",  # dot 5, dots 2356
+    ">": "\u2808\u281c",  # dot 4, dots 345
+    '"': "\u2820\u2836",  # dot 6, dots 2356
+    "[": "\u2828\u2823",  # dots 46, dots 126
+    "]": "\u2828\u281c",  # dots 46, dots 345
+    "\\": "\u2838\u2821",  # dots 456, dots 16
+    "^": "\u2808\u2822",  # dot 4, dots 26
+    "_": "\u2828\u2824",  # dots 46, dots 36
+    "`": "\u2828\u2821",  # dots 46, dots 16
+    "{": "\u2838\u2823",  # dots 456, dots 126
+    "}": "\u2838\u281c",  # dots 456, dots 345
+    "|": "\u2838\u2833",  # dots 456, dots 1256
+    "~": "\u2808\u2814",  # dot 4, dots 35
+}
+
 
 async def test_braille_converter_simple_text():
     """Test basic Braille conversion."""
@@ -131,3 +155,49 @@ async def test_braille_converter_punctuation_cells():
             f"{char!r} -> {result.output_text!r} (U+{ord(result.output_text):04X}), "
             f"expected {cell!r} (U+{ord(cell):04X})"
         )
+
+
+@pytest.mark.parametrize("char, expected", sorted(UEB_SYMBOL_CELLS.items()))
+async def test_braille_converter_ascii_symbol_cells(char, expected):
+    """Printable ASCII symbols map to their UEB cells rather than being dropped.
+
+    Regression: '@', '%', '+', '<' and the other unmapped symbols were silently
+    dropped, corrupting the encoded prompt (e.g. "a@b.com" became "ab.com").
+    Cells are pinned against the Unified English Braille symbol definitions.
+    """
+    converter = BrailleConverter()
+
+    result = await converter.convert_async(prompt=char, input_type="text")
+    assert result.output_text == expected
+
+
+@pytest.mark.parametrize(
+    "char",
+    ["\u00e9", "\u4e2d", "\U0001f600", "\n", "\t", "\r"],
+    ids=["e-acute", "cjk", "emoji", "newline", "tab", "carriage-return"],
+)
+async def test_braille_converter_unmapped_characters_pass_through(char):
+    """Characters with no Braille cell survive conversion unchanged."""
+    converter = BrailleConverter()
+
+    result = await converter.convert_async(prompt=f"a{char}b", input_type="text")
+    assert result.output_text == f"\u2801{char}\u2803"
+
+
+@pytest.mark.parametrize(
+    "prompt, expected",
+    [
+        ("hello", "\u2813\u2811\u2807\u2807\u2815"),
+        ("a@b.com", "\u2801\u2808\u2801\u2803\u2832\u2809\u2815\u280d"),
+        ("1+2", "\u283c\u2801\u2810\u2816\u283c\u2803"),
+        ("100%", "\u283c\u2801\u281a\u281a\u2828\u2834"),
+        ("caf\u00e9", "\u2809\u2801\u280b\u00e9"),
+    ],
+    ids=["letters", "email", "digits-around-symbol", "percent", "accented-letter"],
+)
+async def test_braille_converter_exact_output(prompt, expected):
+    """Mapped characters are still encoded, and pass-through does not corrupt number mode."""
+    converter = BrailleConverter()
+
+    result = await converter.convert_async(prompt=prompt, input_type="text")
+    assert result.output_text == expected