Can I certify an AI-based solution for mission-critical systems?

Question

First I read this one . But I would like to expand. Summarizing:

• When developing security-critical systems, the developer must evaluate certain indicators to gain confidence that the system will work properly. This is a kind of mathematical proof with a fairly low degree of accessibility for humans. This is due to accountability, reliability, audit capability, etc.

• On the other hand, at this point, AI is a black box that works very well, but in most cases we don’t have proof (mainly because the item in the box is too complicated to analyze), it is more like statistical validity: We trained the system, and the system worked well for all tests.

So, some questions?

Q1. Are these two vague thoughts currently available?

Q2. Can AI be used for a critical security system and be sure of its performance? Can we have confidence in the deterministic behavior of AI? Any link?

Q3. I think there are already some companies that sell automotive-critical AI-based security systems, for example. How do they manage to certify their products for such a restrictive market?

EDIT

About Q1: thanks to Peter, I realized that, for example, for an automotive example, there are no requirements for complete confidence. ASIL D, the most restrictive level for automotive systems, requires only an upper estimate of the probability of failure. As well as other standards and ISO26262 levels. I would clarify the question:

Q1. Is there any security standard in system design at any level / subcomponent in any field / domain that requires general confidence?

Q2: , - . , AI: ( ).

Q3: , . , .

Answer 1

. , , , , , . - , , ( - ) . ( AI), ( ), . , , , . , , , , .

"", "". , (RTOS) , , , . , .

. , ( , , , ), , . , , .

, AI ? , , , . , , ( , ), , .

, , . , (, , , , ).

, , . , , .

, . , ( ), (.. , , ). , . , - , , , , AI.

, , . . , , , , , . , , , . , ( ). , , - , . , , , , , , - - .

- . , - - . , , . , , (.. ), , , .