Take-Aways (AI)
  • Fede­ral Coun­cil rati­fi­es Coun­cil of Euro­pe AI Con­ven­ti­on, sees pri­va­te sec­tor affec­ted, wants to exami­ne appro­pria­te mea­su­res against fun­da­men­tal rights risks.
  • No direct adop­ti­on of the EU AI Act; sec­to­ral or spe­ci­fic pro­duct safe­ty regu­la­ti­ons remain pos­si­ble, con­cre­te bans still premature.
  • Trai­ning data and copy­right issues will be exami­ned sepa­ra­te­ly; con­sul­ta­ti­on draft and impact assess­ments plan­ned by the end of 2026.

Inter­pel­la­ti­on Mar­ti (25.3199): AI regu­la­ti­on. Is the wea­k­e­ning of the wea­k­e­ning threatening?

Sub­mit­ted text

On Febru­ary 12, the Fede­ral Coun­cil adopted a Inter­pre­ta­ti­ve docu­ment on AI regu­la­ti­on adopted and defi­ned a num­ber of key points. For exam­p­le, it plans to rati­fy the Coun­cil of Euro­pe Con­ven­ti­on and sub­mit a con­sul­ta­ti­on draft by the end of 2026. In this con­text, I would ask the Fede­ral Coun­cil to ans­wer the fol­lo­wing questions:

1 The Coun­cil of Euro­pe Con­ven­ti­on is only bin­ding for sta­te actors in its mini­mum imple­men­ta­ti­on. In an artic­le in the Repu­blic, civil socie­ty orga­nizati­ons rai­sed the accu­sa­ti­on that the Coun­cil of Euro­pe Con­ven­ti­on had been wate­red down under pres­su­re from the USA and that Switz­er­land had con­tri­bu­ted to this wate­ring down. What is the Fede­ral Council’s posi­ti­on on this accu­sa­ti­on? How sen­si­ble was this in view of the risk that the USA would not rati­fy this con­ven­ti­on with bin­ding force?

2) Is the Fede­ral Coun­cil rely­ing only on a mini­mal imple­men­ta­ti­on of the Coun­cil of Euro­pe Con­ven­ti­on or will it seek a more far-rea­ching solu­ti­on that also invol­ves pri­va­te actors?

3. why does the Fede­ral Coun­cil prio­ri­ti­ze the pro­mo­ti­on of inno­va­ti­on and the posi­ti­ve effects for the eco­no­my in its inter­pre­ta­ti­ve report and its media release instead of human rights or the importance of AI for a sus­tainable, cli­ma­te-fri­end­ly society?

4 The Fede­ral Coun­cil has also announ­ced that it will not adopt the EU AI Act. The Coun­cil of Euro­pe Con­ven­ti­on does not pro­vi­de for the pro­hi­bi­ti­on and regu­la­ti­on of cer­tain high-risk appli­ca­ti­ons, but the EU AI Act does. Does the Fede­ral Coun­cil belie­ve that it is not neces­sa­ry to expli­ci­t­ly pro­hi­bit cer­tain appli­ca­ti­ons (e.g. social scoring) and to regu­la­te ris­ky and sen­si­ti­ve appli­ca­ti­ons (e.g. when using pro­tec­ted per­so­nal data) and thus ensu­re pro­duct safety?

5 Why does the Fede­ral Coun­cil gene­ral­ly refrain from regu­la­ting pro­duct safe­ty and liability?

6 Why is the­re no pro­vi­si­on for the use of trai­ning data for AI appli­ca­ti­ons, espe­ci­al­ly sin­ce Swiss copy­right legis­la­ti­on dif­fers from that of the US and Europe?

7 To what ext­ent is the­re legal cer­tain­ty for Swiss com­pa­nies ope­ra­ting in Euro­pe if Switz­er­land does not adopt Euro­pean law?

8 Why should we wait until the end of 2026 for a pro­po­sal to be drawn up?

9 Why do sus­taina­bi­li­ty con­side­ra­ti­ons play no role in the bench­mark paper?

State­ment of the Fede­ral Coun­cil of 14.5.2025

1. switz­er­land has actively par­ti­ci­pa­ted in the nego­tia­ti­ons of the AI Con­ven­ti­on of the Coun­cil of Euro­pe invol­ved. The Fede­ral Coun­cil is of the opi­ni­on that the objec­ti­ves of the nego­tia­ting man­da­te of 16 Sep­tem­ber 2022 have been achie­ved, in par­ti­cu­lar the har­mo­nizati­on of the inter­na­tio­nal legal frame­work in the field of AI. The AI Con­ven­ti­on is a frame­work agree­ment aimed at the con­trac­ting sta­tes. Its pro­vi­si­ons are not direct­ly appli­ca­ble. Nevert­hel­ess, pri­va­te actors are affec­ted in accordance with Art. 3, para. 1, let­ter b of the AI Con­ven­ti­on. Accor­din­gly, sta­tes are obli­ged to address the risks and impacts ari­sing from acti­vi­ties throug­hout the life cycle of AI systems used in the pri­va­te sec­tor in accordance with the objec­ti­ve and pur­po­se of the Con­ven­ti­on. The aim of this for­mu­la­ti­on was to take appro­pria­te account of the spe­cial fea­tures of various legal systems – par­ti­cu­lar­ly tho­se of count­ries out­side Euro­pe – in the area of fun­da­men­tal rights pro­tec­tion and thus to enable a wider ran­ge of count­ries to rati­fy the Con­ven­ti­on. Such par­ti­cu­la­ri­ties are also rele­vant for Switz­er­land. Switz­er­land signed this con­ven­ti­on on March 27, 2025, as have the Euro­pean Uni­on and 13 other sta­tes, inclu­ding Cana­da and Japan, to date.

2 Accor­ding to the AI Con­ven­ti­on, the­re is grea­ter scope for regu­la­ting the pri­va­te sec­tor than the public sec­tor, but the­re must still be Appro­pria­te mea­su­res taken in order to achie­ve the objec­ti­ves of the Con­ven­ti­on. In its decis­i­on, the Fede­ral Coun­cil poin­ted out that the regu­la­ti­ons will pri­ma­ri­ly affect sta­te actors, but that pri­va­te actors will also be held accoun­ta­ble whe­re neces­sa­ry whe­re fun­da­men­tal rights are poten­ti­al­ly affected.

3 The inter­pre­ta­ti­ve docu­ment lists seve­ral equi­va­lent objec­ti­ves that are to be achie­ved with the cho­sen approach. Among them, the pro­tec­tion of fun­da­men­tal rights is expli­ci­t­ly men­tio­ned. On sus­taina­bi­li­ty, see ans­wer to que­sti­on 9.

4. the approach cho­sen by Switz­er­land allo­ws spe­ci­fic mea­su­res to ensu­re pro­duct safe­ty to be exami­ned and, if neces­sa­ry, imple­men­ted. This also inclu­des the pos­si­bi­li­ty of addres­sing the risks of par­ti­cu­lar­ly high-risk appli­ca­ti­ons appro­pria­te­ly – and if neces­sa­ry by ban­ning cer­tain prac­ti­ces. At the pre­sent time, howe­ver pre­ma­tu­re to name con­cre­te mea­su­res.

5 The Fede­ral Coun­cil has deci­ded against imple­men­ting the AI Con­ven­ti­on on the basis of the AI Act and is the­r­e­fo­re oppo­sed to the intro­duc­tion of hori­zon­tal legis­la­ti­on aimed exclu­si­ve­ly at the pro­duct safe­ty of AI systems. Howe­ver, the cho­sen approach does not rule out the pos­si­bi­li­ty of sec­tor-spe­ci­fic tech­ni­cal pro­duct regu­la­ti­ons being issued or the spe­cial fea­tures of AI systems being taken into account in any future revi­si­ons of exi­sting spe­cial legislation.

6 The Fede­ral Coun­cil does not rule out requi­re­ments for the use of trai­ning data for AI systems. Howe­ver, this topic does not fall direct­ly within the scope of the AI Con­ven­ti­on. The Deal­ing with copy­right-pro­tec­ted con­tent in trai­ning data is the­r­e­fo­re being exami­ned as part of sepa­ra­te workpar­al­lel to the pre­pa­ra­ti­on of the con­sul­ta­ti­on draft sche­du­led for the end of 2026. In this con­text, the Fede­ral Coun­cil would like to point out that it will not accept the Moti­on Gös­si 24.4596 “Bet­ter pro­tec­tion of intellec­tu­al pro­per­ty against AI misu­se” which calls for an in-depth exami­na­ti­on of this topic.
7 Swiss com­pa­nies that place pro­ducts with AI com­pon­ents on the EU inter­nal mar­ket or ope­ra­te AI systems who­se results are used in the EU must com­ply with the requi­re­ments of the AI Act. Two steps would be neces­sa­ry to faci­li­ta­te access to the EU inter­nal mar­ket: first­ly, the intro­duc­tion of equi­va­lent rules in Swiss law, and second­ly, an amend­ment to the Swiss-EU agree­ment on the mutu­al reco­gni­ti­on of con­for­mi­ty assess­ments (Mutu­al Reco­gni­ti­on Agree­ment, MRA; SR 0.946.526.81). Adop­ti­on of the AI Act by Switz­er­land alo­ne would not gua­ran­tee easier mar­ket access.

8. the dead­lines have been esti­ma­ted accor­ding to the time requi­red for the pre­pa­ra­ti­on of a preli­mi­na­ry draft for a law of this magnitu­de. In addi­ti­on, a regu­la­to­ry impact assess­ment (RIA) and, if neces­sa­ry, a data pro­tec­tion impact assess­ment (DPIA) will be car­ri­ed out for this proposal.

9 Alt­hough que­sti­ons rela­ting to sus­taina­bi­li­ty were not inclu­ded in the inter­pre­ta­ti­ve docu­ment, the Fede­ral Coun­cil has nevert­hel­ess deci­ded that the­se aspects should be exami­ned in grea­ter depth in sepa­ra­te work. In par­ti­cu­lar, it sup­port­ed the adop­ti­on of the Po. Fivaz 24.4679 “Does the power con­sump­ti­on of arti­fi­ci­al intel­li­gence pose a thre­at to the Ener­gy Stra­tegy 2050?” applied for.