MarIus

4.2.1 The mechanism in the AIA Article 2(2), and the new MED Article 8(5)

If ANS are classified as high-risk AI systems under the AIA Article 6(1) on the basis that they are covered by the MED, Article 2(2) prevents the requirements for high-risk AI systems to apply via Article 6(1).^{(1) Thesis Section 3.3.3.} However, Article 2(2) decides that other provisions of the AIA apply in these situations. For ANS, Article 105 is the provision of relevance.

The AIA Article 105 amends the MED Article 8 with a new (fifth) paragraph. The new paragraph requires the European Commission to "take into account" the system-oriented requirements for high-risk AI systems in the AIA Chapter III, Section 2, when performing its activities pursuant to the MED Article 8(1), and when exercising its authority pursuant to the MED Articles 8(2) and (3). The two latter provisions are discussed below.^{(2) Thesis Section 4.2.4.} The MED Article 8(1) does not directly concern EU-regulation of ANS, but I will touch upon the provision when considering the "Brussels Effect".^{(3) Thesis Section 4.3.4.}

As determined by the AIA Article 2(2), Article 8(5) of the MED applies only to AI systems that are classified as high-risk pursuant to the AIA Article 6(1).^{(4) Considered with respect to ANS in Section 4.2.2 below.} Also, Article 8(5) limits its scope to AI systems that are "safety components within the meaning" of the AIA.^{(5) Considered with respect to ANS in Section 4.2.3 below.}

4.2.2 The ANS must be classified as high-risk pursuant to the AIA Article 6(1)

Pending clarification from the European Commission, the question of whether ANS constitute high-risk AI systems under the AIA Article 6(1) remains a matter of interpretation.^{(6) Thesis Section 3.3.2.} The provision demands that ANS are "covered" and "required to undergo a third-party conformity assessment" by the MED.^{(7) Thesis Section 3.3.3.}

The question of whether ANS fall within the scope of the MED has already been addressed in this thesis.^{(8) Thesis Section 3.4.2.2.} Here, however, the potential geographical allocation of components of ANS in different locations is brought into focus from a slightly different angle.^{(9) Thesis Sections 2.1.2 (actual premises), 3.4.2.2 and 3.4.2.3 (significant in the research question of whether the requirements for high-risk AI systems apply to ANS).} In order for the requirements for high-risk AI systems to apply to ANS, this presupposes that a sufficient number of components are located either on non-EU-ships or on shore, such that those components constitute an "AI system" escaping the lex specialis function of the MED.^{(10) Thesis Section 3.4.2.} Conversely, the premise for a potential indirect impact via the MED Article 8(5) is that a sufficient portion of components from an autonomous navigation system are located on board an EU-ship, such that they collectively constitute an "AI system" under the AIA that can be "covered" under the MED pursuant to the AIA Article 6(1). Given that most elements of the definition of an AI system relate to the autonomous capabilities of the AI system, the decisive factor will likely be whether the "brain" of the autonomous navigation system is situated on board the ship.^{(11) Thesis Section 3.2.1.}

If an autonomous navigation system is already covered by the MED, this would already imply a recognition that the components of the system require flag State approval pursuant to Article 3(1) of the MED. Such approval would certainly constitute a "third-party conformity assessment" within the meaning of Article 6(1) of the AIA. The relevant flag State authorities will certainly perform "assessment activities, including testing, certification and inspection", aligned with the definition of a "conformity assessment body" in the AIA Article 3(21).

4.2.3 The ANS must be "safety components" within the meaning of the AIA

A "safety component" is legally defined in the AIA Article 3(14) as a component of a "product". The component must either fulfill "a safety function" for that product or "the failure or malfunctioning of which endangers the health and safety of persons or property".

It does not seem problematic to consider an autonomous ship as a "product" that an autonomous navigation system can be a component of. Nothing in the AIA suggests otherwise, and the concept of a product remains vague throughout the Recitals and Explanatory Memorandum.^{(12) Well demonstrated by the AIA Recitals para. 12.} It might be intentionally vague, designed to encompass a wide range of items in which AI systems can become a component. Moreover, it does not seem problematic that components of ANS might be partially situated outside of the products (the MASS) themselves.^{(13) Thesis Section 2.1.2.} The Recitals emphasize that it is not of definitional significance whether AI systems are "embedded" or "non-embedded" in the products.^{(14) The AIA Recitals para. 12.}

Furthermore, ANS will fulfil safety-enhancing functions for MASS, while their failure or malfunction may endanger the surroundings of the ships, and even disrupt supply chains.^{(15) Thesis Section 3.3.4.3.} In summary, ANS are likely "safety components within the meaning" of the AIA, as required by the MED Article 8(5).

4.2.4 Grounds to believe that the AIA will in fact have an impact through the new MED Article 8(5)