**AES** is a ‘symmetric block cipher’ for encrypting texts which can be decrypted with the original
encryption key.

For many purposes, a simpler encryption algorithm such as TEA is perfectly
adequate – but if you suspect the world’s best cryptographic minds, and a few million dollars of
computing resource, might be attempting to crack your security, then AES,
based on the *Rijndael* algorithm, is the tightest security currently available (approved by
the US government for classified information up to ‘Secret’ – and in in 192 or 256 key lengths, up
to ‘Top Secret’). AES was adopted by NIST in 2001 as
FIPS-197, and is the replacement for DES which was
withdrawn in 2005.

I developed this JavaScript implementation to to illustrate the original AES standard (NIST FIPS-197) as closely as possible. It is intended as an introduction for people seeking to learn something about implementing encryption, not an authoritative implementation – cryptography experts will already know more than I present here. The emphasis is on transparency and fidelity to the standard rather than efficiency.

This script also includes a wrapper function which implements AES in the ‘Counter’ mode of operation (specified in NIST SP 800-38A) to encrypt arbitrary texts – many descriptions of AES limit themselves to the Cipher routine itself, and don’t consider how it can be used to encrypt texts.

Much of the Rijndael algorithm is based on arithmetic on a *finite
field*, or *Galois field* (after the mathematician). Regular arithmetic works on
an infinite range of numbers – keep doubling a number and it will get ever bigger. Arithmetic
in a finite field is limited to numbers within that field. The Rijndael algorithm works in GF(2^{8}),
in which arithmetic results can always be stored within one byte – which is pretty convenient
for computers. I can’t begin to understand the maths (considering
that addition and subtraction are the same thing – an xor operation
– and multiplication is performed ‘modulo an irreducible polynomial’: doubling 0x80 in GF(2^{8})
gives 0x1b).

The Rijndael algorithm lends itself to widely differing implementations, since the maths can be either coded directly, or pre-computed as lookup tables – directly parallel to using log tables for arithmetic. Different implementations can have varying pay-offs between speed, complexity, and storage requirements. Some may barely resemble each other. In this implementation, I have followed the standard closely; as per the standard, I have used a lookup table (‘S-box’) to implement the multiplicative inverse (i.e. 1/x) within a finite field (used for the SubBytes transformation), but other calculations are made directly rather than being pre-computed.

If you want to convince yourself that the Cipher function is working properly internally (and you
should!), NIST provide test vectors for AES (appendix C.1 of the standard). Click

and the cipher output block should be

- 128-bit:
*69 c4 e0 d8 6a 7b 04 30 d8 cd b7 80 70 b4 c5 5a* - 192-bit:
*dd a9 7c a4 86 4c df e0 6e af 70 a0 ec 0d 71 91* - 256-bit:
*8e a2 b7 ca 51 67 45 bf ea fc 49 90 4b 49 60 89*

(In counter mode, a text could decrypt correctly even if the cipher routine was flawed).

The Inverse Cipher is largely a mirror of the Cipher routine, with parallel functions for Cipher, SubBytes and ShiftRows. The MixColumns routine is slightly more complex in the inverse. I have not implemented the inverse cipher here as it is not required in counter mode.

**Counter mode of operation**: the AES standard concerns itself with numeric or binary data (Rijndael,
along with most other encryption algorithms, works on a fixed-size block of numbers – in the case
of AES, each block is 128 bits or 16 bytes).

In order to make use of it to encrypt real things (such as texts), it has to be used within a certain ‘mode of operation’. This is the interface between text or files, and the purely numerical encryption algorithm. See NIST Special Publication SP800-38A for more details and test vectors.

The simplest mode of operation (‘electronic codebook’) encrypts a text block-by-block – but since the same block of plaintext will always generate the same block of ciphertext, this can leave too many clues for attackers.

In the ‘counter
mode’ used in this implementation, a counter which changes with each block is first encrypted,
and the result is bitwise xor’d with the plaintext block to
get the ciphertext block (so the plaintext is not actually directly encrypted). A unique ‘nonce’
is incorporated in the counter to ensure different ciphertexts are always generated from the
same plaintext every time it is encrypted; this number is stored at the head of the ciphertext
to enable decryption. A combination of seconds since 1 Jan 1970, a millisecond-timestamp, and
a sub-millisecond random number gives a very effective nonce. (To resist cryptographic attacks,
the nonce does not need to be secret or unpredictable, but it is imperative that it is unique).
In this implementation, the initial block holds the nonce in the first 8 bytes, and the block
count in the second 8 bytes. Since JavaScript can represent integers up to 2^{53},
this allows a message size up to 2^{57} (c.
10^{17})
bytes – unlikely to be a limitation! Note that the nonce in counter mode is the equivalent
of the initialisation vector (IV)
in other modes of operation.

A curious quality of counter mode is that decryption also uses the cipher algorithm rather than the inverse-cipher algorithm. Though simple to implement, it has been established to be very secure.

Encrypting texts or files require not just the mode of operation. When implementing AES, you have to consider

- mode of operation; here the Counter (ctr) mode of operation – both simple to implement, and very secure
- conversion of text (including multi-byte Unicode texts) to binary/numeric data; here multi-byte Unicode characters are converted to UTF8, then the numeric character codes are used to pass to the cipher routine
- conversion of encrypted data to values which can be stored or transmitted without problem; here the binary encrypted texts are encoded in Base64, which is a very safe 7-bit encoding with no control codes or other troublesome characters.

The **key** in this script is obtained by applying the Cipher routine to encrypt the first
16/24/32 characters of the password (for 128-/192-/256-bit keys) to make the key. This is a
convenient way to obtain a secure key within an entirely self-contained script (in a production
environment, as opposed to this essentially tutorial code, the key might be generated
as a hash, e.g. simply `key = Sha256(password)`

).
In more detail, the supplied password is converted to to UTF-8 (to be byte-safe), then
the first 16/24/32 characters are converted to bytes. The resulting pwBytes is used as a seed
for the Aes.keyExpansion() which is then used as the key to encrypt pwBytes with Aes.cipher().
Examples of keys generated in this way from (unrealistically) simple passwords:

‘a’ (U+0061): | pwBytes = | 61 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |

key = | 60 84 dd 49 14 7b 5d 05 7a e3 f8 81 b9 0e e7 dd | |

‘b’ (U+0062): | pwBytes = | 62 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |

key = | b4 1a 83 4f da 4b aa 41 76 62 be d6 2c 66 83 6d | |

‘☺’ (U+263a): | pwBytes = | e2 98 ba 00 00 00 00 00 00 00 00 00 00 00 00 00 |

key = | d1 0c cd fd 44 45 54 ef 59 aa f8 dc 78 8e 9a 7c |

Even with a single bit difference between two passwords (‘a’ and ‘b’), the key is entirely different.

**Usage**: this implementation would be invoked as follows:

var password = 'L0ck it up saf3'; var plaintext = 'pssst ... đon’t tell anyøne!'; var ciphertext = Aes.Ctr.encrypt(plaintext, password, 256); var origtext = Aes.Ctr.decrypt(ciphertext, password, 256);

Note that there are no standards for data storage formats of AES encryption mode wrapper functions, so this is unlikely to inter-operate with standard library functions, though there is a matching PHP version.

It is also quite simple to encrypt files, by using the HTML5 FileReader & Blob objects: (not available on IE9-)

Note that Aes.Ctr.encrypt expects a string: as binary files may include invalid Unicode sequences if treated as strings, I treat the file contents as a byte-stream, converting it to single-byte characters before passing it to Aes.Ctr.encrypt.

*Does it make sense to implement AES in JavaScript?* This is really intended as a
reference implementation to help understand the AES standard, but sometimes JavaScript can be
used for real-world cryptographic applications (particularly web-based ones). Mostly, a JavaScript
implementation such as this can provide an easy starting-point for implementation in other languages.
Though I still think that TEA is generally good enough for simple applications,
and a great deal simpler to use (as well as significantly faster, according to
tests by Tom Doan, thanks Tom).

*In other languages:* I’ve developed a PHP version which directly
mirrors this JavaScript version; it differs in that PHP has Base64 encoding and UTF-8 encoding built-in,
and has no unsigned-right-shift operator(!), but is otherwise a straightforward port. In other languages,
be sure to use 64-bit integers/longs, either unsigned or with unsigned right-shift operators; you
may need to take into consideration the way different languages handle bitwise ops, and of course
standard issues such as array handling and strict typing. I’m not aware of any other issues.

I’m not familiar with Python, but there is a Python version available at wiki.birth-online.de/snippets/python/aes-rijndael.

*Speed:* as mentioned, this is not an optimised implementation – using Chrome on a low-to-middling
2014 machine (Core-i5), this processes around 1MB/sec [still some 100× faster than back in 2008!).

For more information, have a look at

- Daemen & Rijnael’s AES proposal for Rijndael Block Cipher
- Wikipedia article
- article from John Savard’s Cryptographic Compendium

For some security applications, a cryptographic hash is more appropriate than encryption – if you are interested in a hash function, see my implementations of SHA-1 and SHA-256.

*October 2009*: I have updated the formulation of these scripts to use JavaScript namespaces for better
encapsulation of function names.

See below for the source code of the JavaScript implementation, also available on GitHub. §ection numbers relate the code back to sections in the standard.

With its untyped C-style syntax, JavaScript reads remarkably close to pseudo-code: exposing the algorithms with a minimum of syntactic distractions. These functions should be simple to translate into other languages if required, though can also be used as-is in browsers and Node.js.

For convenience & clarity, I have extended the base JavaScript String object with utf8Encode(), utf8Decode(), base64Encode(), base64Decode() methods: I don’t see great likelihood of conflicts.

I offer these scripts for free use and adaptation to balance my debt to the open-source info-verse. You are welcome to re-use these scripts [under an MIT licence, without any warranty express or implied] provided solely that you retain my copyright notice and a link to this page.

If you would like to show your appreciation and support continued development of these scripts, I would most gratefully accept donations.

If you have any queries or find any problems, contact me at ku.oc.epyt-elbavom@cne-stpircs.

*© 2005-2014 Chris Veness*