Skip to content

BSI AIS 20/31 and Common Criteria, the European route

AIS 20/31 is the BSI evaluation methodology for random number generators used in Common Criteria evaluations in Europe. It classifies generators by functionality class — deterministic, physical or hybrid — and matters commercially in defence, government and regulated infrastructure procurement.

The European route

Where NIST publishes SP 800-90 and validates against it through CAVP and ESV, the German federal office BSI publishes AIS 20 and AIS 31 as evaluation methodology used within Common Criteria. The goals overlap substantially; the structure and vocabulary do not.

A product sold into both markets will often carry evidence under both, which is a real cost and a reason to ask vendors which they hold rather than assuming one implies the other.

Functionality classes

  • DRG classes cover deterministic generators, graded by how much is required of the seed and of forward and backward secrecy.
  • PTG classes cover physical true random number generators. PTG.2 is the class most often cited for a physical source with a stochastic model and continuous testing.
  • NTG / hybrid classes cover non-physical and combined constructions, where a deterministic mechanism is continuously reseeded from a physical source.

What the methodology emphasises

A stochastic model of the noise source. That is the same requirement SP 800-90B makes, expressed differently, and it reflects the same underlying judgement: an entropy claim without a model of the process behind it cannot be evaluated.

The evaluation also expects online testing appropriate to the class, which again converges with the NIST position that a physical source has to be checked in service and not only at evaluation.

Where it decides a purchase

In European defence and government procurement, in regulated infrastructure, and wherever Common Criteria is the framework a customer already uses. In those settings the question is not whether a source is good but whether the evidence is in the form the process consumes.

The two schemes ask compatible questions, and a source with a written model and runtime health evidence is in a position to answer both. The buyer-side questions are largely scheme-independent for that reason.