Demonstration · red team

ULTIMA, the adversarial workbench

Deep Blue Krypto does not ask you to trust that an implementation is sound. We attack it ourselves, on purpose, and gate the fix so the same weakness cannot return. ULTIMA is the purple-team loop: it finds, ZAP hardens, and a resilience gate locks the result.

findharden gatedefensive only

Demonstration. A walkthrough of the ULTIMA method. No live attack is executed from this page.

The loop

Find, harden, gate

Find
ULTIMA attacks the implementation
Harden
ZAP fixes the root cause
Gate
a regression gate locks it

A finding is closed only when a gate proves the fix and would catch the regression, not when someone says it is fixed.

What it looks for

Implementation security, not just algorithm choice

Choosing a NIST parameter set is the easy part. The failures that matter live in the implementation: timing side channels, decryption-failure oracles, missing consistency checks, and trust stories an artifact cannot actually support.

Constant-time

Timing and side channels

Secret-dependent branches and memory access are measured against a constant-time gate (Delta and dudect). A secret must not be readable from how long an operation takes.

Oracles

Decryption-failure oracles

FIPS 203 implicit rejection is preserved: a bad ciphertext must not be distinguishable from a good one before the authentication tag, or the decision leaks.

Consistency

Pairwise and verify-after-sign

Keys are checked for pairwise consistency and withheld on failure; ML-DSA signatures are verified after signing before they are relied upon.

Trust stories

Claims the artifact supports

A label such as "post-quantum" or "forward-secret" is allowed only where the artifact actually provides the property. The mismatch is the finding.

Run an attack

Execute from this page

The live ULTIMA workbench runs inside the sovereign edge against authorized, in-scope targets only. No attack is launched from this public page.

Run timing probe (not enabled) Run oracle probe (not enabled)

FLAG: not executable here, by design. ULTIMA operates on our own stack or an explicitly authorized target, inside the sovereign edge, with results graded and disclosure owner-gated. It is defensive and legitimate-use only. These controls are intentionally disabled in this public demonstration; nothing on this page attacks anything.

Hold us to it

Ask how we attack our own implementations before we ship them.