FortiBleed Showcases Poor Practice
Why did a major cybersecurity company use SHA-256 to hash passwords?
Let me start with a bit of theory on password hashing that we teach our students.
Never use SHA-256 for password hashing!
If I have learned anything about hashing passwords, I should not use a fast hashing method, as it can often be attacked using a standard Hashcat dictionary attack. SHA-256, for example, should not be used, as it has high throughput for Hashcat or AI-based methods, especially when using a GPU infrastructure. With this, there is no looping of the hash password, and which disables the parallel processing of the hashed value against standard dictionaries. So, although SHA-256 is almost impossible to break, many users will use fairly standard patterns for passwords (e.g., replacing an ‘s’ with a ‘5’, and bringing together standard words with one or more numbers and characters(typically a ‘!”) at the end.
I repeat … do not use SHA-256 to hash passwords. For this, we should use the GPU-busting methods of PBKDF2, scrypt or bcrypt. These methods hash for a number of rounds for the hashing process, and which considerably slows down the hashing process - and make it difficult for GPUs to crack. In fact, to properly disable GPU cracking, we should use a method such as Argon2.
Bcrypt, for example, has a field that defines the number of rounds that we hash the password, and where a value of 6 identifies that we hash for 2⁶ rounds:
If you are interested, here are the enhanced password hashing methods in action:
FortiBleed
And, so, it is thought that FortiBleed affects around half of all the Fortinet firewalls (and over 1,000 organisations) and involves a leakage of hashed VPN passwords [here]. Why are these Fortinet firewalls easier to compromise? Well, some of the firewalls seem to have been using older hashing methods, and which can be cracked with GPU approaches.
With this, a while back, Fortinet replaced SHA-256 salted passwords with the improved PBKDF2 method, but where existing administrator passwords remain stored as SHA-256 hashes. On the first login of an administrator into the firewall, a PBKDF2 upgrade is applied. In some cases, this might not have happened. Another strange feature is that even 20-character passwords have been successfully compromised, as they have known plaintext mappings. Along with this, threat actors are already trying to crack the discovered hashes and salt values using Hashtopolis — a dedicated 45-GPU cluster: https://hashtopolis.eu/
SOCRadar [here] report that a database of over 86,644 working credentials has been collected from Internet-facing Fortinet devices. This data has been collected from February 2026 and used automated scanning and configuration file exfiltration.
The patches involve [here]:
- CVE-2026–248589.4 FortiOS, FortiManager, FortiAnalyzer, FortiWeb, FortiProxyFortiCloud SSO bypass.
- CVE-2025–597189.8 FortiOS, FortiWeb, FortiProxy, FortiSwitch ManagerSAML SSO authentication bypass via crafted SAML message.
- CVE-2025–597199.8 FortiOS, FortiWeb, FortiProxy, FortiSwitch ManagerCompanion SAML SSO bypass.
- CVE-2026–356169.1 FortiClient EMS 7.4.5, 7.4.6Unauthenticated API bypass.
- CVE-2026–216438.6 FortiClient EMS 7.4.4 Unauthenticated SQL injection.
- CVE-2022–406849.8 FortiOS, FortiProxy, FortiSwitchManagerHistorical.
