John the Ripper: No Password Hashes Loaded (All 8 Causes)

John refuses to load your hash file. The eight real causes in frequency order, the build trap that hits most Kali users, and a five-command diagnostic that…

John the Ripper prints No password hashes loaded and stops before trying a single guess. The line above it usually says Using default input encoding: UTF-8, which makes it look like an encoding problem. It almost never is.

The message means John read your file, tried to recognise at least one line as a hash whose format it supports, and failed on every line. There are eight reasons that happens, and the popular answers online cover two of them — one of which comes from a page that tells you to store passwords in plain text instead of a hash, so start elsewhere.

Below, the causes are ordered by how often they actually occur, with the command that proves each one.

Read the message precisely: three different sentences

OutputWhat it meansWhere to go
No password hashes loaded (see FAQ)Zero lines were recognised. A whole-file problem.Causes 1–8 below
No password hashes left to crack (see FAQ)Every hash in the file is already in john.pot. Nothing is broken.Cause 1 — this is the good one
Warning: invalid ciphertext ignoredJohn saw a hash, but it did not match the format you forced.Cause 3
Loaded 0 password hashes with 0 different saltsSame as the first case, with the count spelled out.Start at cause 1

Everything below assumes you are in the directory that holds the file. Run ls -l once; a surprising number of these turn out to be the wrong file in the current directory.

1. You already cracked it

John stores every recovered password in ~/.john/john.pot and refuses to re-crack what it already knows. Run the same command twice and the second run fails with this error, which reads like a breakage and is not.

# Is it already cracked?
john --show hashes.txt

# Proof: the pot file already holds this hash
ls -l ~/.john/john.pot
grep -c '\$' ~/.john/john.pot

If --show lists the plaintext, you are done — there is nothing to load. To deliberately re-crack the same hashes (to benchmark a wordlist, for example), move the pot file aside first and keep a copy:

mv ~/.john/john.pot ~/.john/john.pot.bak
john --wordlist=rockyou.txt hashes.txt
mv ~/.john/john.pot.bak ~/.john/john.pot

2. Your build does not have the format

This is the single biggest trap on Kali and Debian, and it is invisible: john from the distribution package is the core build, which supports a short list of formats — descrypt, bcrypt, md5crypt, LM and a few others. Raw MD5, raw SHA-256, NTLM and the *2john converters all live in jumbo. If the format is not compiled in, John cannot load the line no matter how correct it is.

# Does this build know the format at all?
john --list=formats | tr ',' '\n' | grep -i -E 'raw-md5|raw-sha256|nt$'

# If that prints nothing, you have the core build. Check the version banner:
john --version
john --list=build-info | head -20

On Debian and Ubuntu the jumbo packages are separate: apt install john-data still gives you core on some releases, so the reliable path is the upstream bleeding-jumbo build or the Kali package john after kali-tweaks pulls jumbo. On macOS, brew install john-jumbo — plain brew install john has historically been core.

# Rebuild jumbo from source (Linux/macOS), about three minutes
git clone --depth 1 https://github.com/openwall/john.git
cd john/src && ./configure && make -sj4
../run/john --list=formats | head -c 200

3. --format names the wrong algorithm

Forcing a format is the right move when auto-detection guesses wrong, and the wrong move when you are guessing. --format=md5crypt on a bare 32-character MD5 hash is a contradiction: md5crypt hashes start with $1$, plain MD5 does not, so John rejects every line and reports zero loaded.

# A bare 32-hex hash is raw MD5, not md5crypt
john --format=raw-md5 --wordlist=rockyou.txt hashes.txt

# Not sure? Drop the flag entirely and let it detect
john --wordlist=rockyou.txt hashes.txt

# Look up the exact name this build uses
john --list=formats | tr ',' '\n' | grep -i md5

The common confusions: raw-md5 vs md5crypt, nt vs lm, sha512crypt vs raw-sha512, and bcrypt vs bcrypt-opencl. If --format is on the command line and nothing loads, remove it before changing anything else.

4. A shadowed /etc/passwd

On any modern Unix, /etc/passwd contains an x where the hash used to be. John is not wrong to reject those lines, they contain no hash. You need the shadow file and you need them merged into one file with the username kept intact.

# Wrong: no hashes in this file at all
john /etc/passwd

# Right: merge passwd + shadow, then crack
sudo unshadow /etc/passwd /etc/shadow > unshadowed.txt
john --wordlist=rockyou.txt unshadowed.txt

Check the merge before you run it: if unshadowed.txt has no $ characters, the merge silently produced nothing usable (common when reading a container image or a backup, where the shadow file is a stub).

grep -c '\$' unshadowed.txt
awk -F: 'NF<2' unshadowed.txt | head

5. A PWDUMP file with only NTLM hashes

Secretsdump, mimikatz and Impacket produce user:rid:lmhash:nthash:::. When the LM half is empty (aad3b435b51404eeaad3b435b51404ee is the empty-LM placeholder), John's default behaviour on a PWDUMP file is to look at LM, find nothing crackable, and report zero loaded. The NTLM hashes are right there in field four.

# Force the NTLM half
john --format=nt --wordlist=rockyou.txt dump.txt
john --format=nt --show dump.txt

# Or reduce the file to bare NTLM hashes and let detection work
awk -F: '{print $4}' dump.txt | grep -v '^$' > nt.txt
john --format=raw-md5 --wordlist=rockyou.txt nt.txt   # NTLM is raw MD5 of UTF-16LE

Watch the ordering trap here: NTLM is not "raw-md5 of the password", it is MD5 of the UTF-16LE encoding, which is why it needs --format=nt and not raw-md5 for a real crack. Use the field-four version and force nt.

6. The file is not plain ASCII text

John loads text files it can read as UTF-8 with one hash per line. PowerShell's Out-File writes UTF-16LE by default, Notepad has historically added a BOM, and anything copied out of a Windows terminal can arrive with carriage returns glued to the end of each line.

file hashes.txt
# hashes.txt: Unicode text, UTF-16, little-endian     <-- the problem

iconv -f UTF-16 -t UTF-8 hashes.txt > hashes-utf8.txt
sed -i 's/\r$//' hashes-utf8.txt
john --wordlist=rockyou.txt hashes-utf8.txt

This produces Token length exception in hashcat and a quiet zero-loaded in John, from the same cause. The cat -A check from the hashcat token length page applies here unchanged: ^M at the end of each line is your answer.

7. The converter produced something John does not read

Running zip2john archive.zip > hash.txt and getting nothing back is common enough to have its own sequence. Three things go wrong, in this order:

  • zip2john is not on the PATH. In a jumbo build the binaries live in run/, not /usr/bin. find / -name 'zip2john' 2>/dev/null settles it.
  • The output went to stderr, not stdout. Some converters write the hash and a warning; if you redirected only stdout you may have an empty file. Check wc -c hash.txt — zero bytes is the tell.
  • It is a 7z or a split archive. zip2john does not handle every ZIP variant; 7-Zip archives need 7z2john.pl, and a split archive has to be rejoined first.
wc -c hash.txt                  # zero means the converter never wrote a hash
head -c 120 hash.txt            # should show a $-prefixed structure or filename:hash
john hash.txt --format=zip      # force it if auto-detection is unsure

8. The line is not what you think it is

Two clipboard accidents account for most of the rest. A hash pasted out of a web page or a chat window loses characters at a line wrap, or comes in with its username still attached (admin:$1$...), or with surrounding quotes from a shell command.

# Length per line: for one hash type, every line should be identical
awk '{print length($0)}' hashes.txt | sort -n | uniq -c

# Leading username or stray whitespace
awk -F: '{print NF}' hashes.txt | sort -u
cat -A hashes.txt | head -3

The five-command diagnostic

Run these in order. Each one rules out a cause, so you stop guessing after the first answer that looks wrong.

john --version                                       # 1. jumbo or core?
john --list=formats | grep -ci md5                   # 2. does it know the algorithm?
john --show hashes.txt; ls -l ~/.john/john.pot       # 3. already cracked?
file hashes.txt; cat -A hashes.txt | head -3         # 4. encoding, line endings
awk '{print length($0)}' hashes.txt | sort -u | wc -l  # 5. one consistent length?

If all five are clean and it still loads nothing, the file is almost certainly in a format this build does not support — at which point the fastest path is a jumbo build on the same machine, not another flag.

Confirm the fix

A working run prints the load count and then a speed:

Loaded 1 password hash (Raw-MD5 [MD5 32/64 X2])
Press 'q' or Ctrl-C to abort, almost any other key for status
0g 0:00:00:02 3/3 0g/s 1234Kp/s ...

A non-zero speed means the hash loaded and John is guessing. Zero recovered after a full wordlist is a different problem with a different answer: the password is not in that wordlist, or it is in a format the mode mismatch hides. That is a candidate-set problem, not a loading problem.

Frequently asked

Why does john say no password hashes loaded when my hash is definitely valid?

Because validity is not the question. The line has to match a format the build supports and a shape John can recognise. Check john --list=formats for the algorithm first, then the encoding of the file. A correct hash in a format the build lacks produces exactly this message.

Does john load one hash per line?

For raw hashes, yes, one per line and nothing else on the line. The /etc/passwd and PWDUMP layouts are the exception, and they need unshadow or --format respectively.

What is the difference between this error and "no password hashes left to crack"?

The second one means every line is already in john.pot. Nothing is wrong. Print the answers with john --show.

Can I crack NTLM hashes with the core build?

No. NTLM needs jumbo. If you cannot install jumbo, this is the moment to use Hashcat instead — hashcat -m 1000 handles the same field-four extraction, and the comparison page covers which tool to start with.

Where does the wordlist go?

--wordlist=rockyou.txt before the hash file. On Kali it is at /usr/share/wordlists/rockyou.txt, and on Debian it may be compressed as rockyou.txt.gz and need decompressing first.

On hashes you own or are authorised to test. Hashes lifted from a breach dump are someone else's data, and using a recovered password to sign in anywhere is a criminal act in most jurisdictions, regardless of how the hash was obtained.

Where this fits

Loading is the boring half of cracking and the half that stops most sessions. Once the file loads, the rest is candidate selection: wordlist, then rules, then masks. Practice the whole path on the John cheat sheet, read the tool guide for the converters, and run your own hash strings through the Hash & JWT Inspector to see what a format looks like before you ask a tool to guess.

See also: Hashcat "No Hashes Loaded" — the same load-time failure family in hashcat, with its token-length, hash-mode and salt-value variants.


Keep reading

Related articles

Browse all articles