McEliece cryptosystem is the first public-key cryptosystem based on linear error-correcting codes. Although a code with an efficient bounded distance decoding algorithm is chosen as the secret key in this cryptosystem, not knowing the secret code and its decoding algorithm faced the attacker with the problem of decoding a random-looking linear code. Moreover, it is well known that the known efficient bounded distance decoding algorithm of the families of codes proposed for code-based cryptography (like Reed-Solomon codes, Goppa codes, alternant codes or algebraic geometry codes) can be described using error correcting pairs (ECP). That means that, the McEliece cryptosystem is not based on the intractability of bounded distance decoding but on the problem of retrieving an error-correcting pair from a random linear code. The aim of this article is to propose the class of codes with a t-ECP whose error-correcting pair is not easily reconstructed from the single knowledge of a generator matrix.