Take-Aways (AI)
  • Expert opi­ni­on recom­mends not sto­ring sen­si­ti­ve per­so­nal data in Micro­soft clouds due to loss of con­trol and access by US authorities.
  • Encryp­ti­on (effec­ti­ve against pro­vi­der access) or local/alternative hosting in non-US data cen­ters pro­po­sed as miti­ga­ting measures.
  • Cri­ti­cism: The aut­hors’ inter­pre­ta­ti­on of the US SCA is dis­pu­ted; risk assess­ment (pro­ba­bi­li­ty and con­se­quen­ces of access) remains central.

Dr. Phil­ip Glass and Prof. Dr. Mar­kus Sche­fer have been com­mis­sio­ned by egov­part­ner (a coope­ra­ti­on orga­nizati­on of muni­ci­pa­li­ties, cities and the Can­ton of Zurich) has drawn up an expert opi­ni­on on the Use of Micro­soft 365 by public bodies in Switz­er­land in com­pli­ance with fun­da­men­tal rights (using the exam­p­le of the Can­ton of Zurich). The expert opi­ni­on was in the Jus­let­ter IT of Decem­ber 20, 2023 and is available as Edi­ti­ons Web­law, Bern 2023. It poses the question,

how the muni­ci­pa­li­ties in the Can­ton of Zurich can use cloud ser­vices (in par­ti­cu­lar M365) in a con­sti­tu­tio­nal and data pro­tec­tion-com­pli­ant manner.

The report was made available to the muni­ci­pa­li­ties of the Can­ton of Zurich pri­or to publi­ca­ti­on and has been allo­wed to be pas­sed on to third par­ties sin­ce the begin­ning of Octo­ber 2023. The aut­hors recei­ved feed­back based on this; they address this in an adden­dum to the report.

The expert opi­ni­on can also can be down­loa­ded from David Rosenthal’s web­site. David Rosen­thal alre­a­dy wro­te about this on Novem­ber 10, 2023 Comm­ents writ­ten on the expert opi­ni­on; see below.

Results of the expert opinion

The experts essen­ti­al­ly come to the fol­lo­wing conclusions:

  • Art. 13 para. 2 BV pro­tects against the misu­se of data as a mini­mum gua­ran­tee of lawful data pro­ce­s­sing and pro­vi­des a legal­ly pro­tec­ted inte­rest in cor­rect data pro­ce­s­sing in accordance with the rule of law. In addi­ti­on, Art. 13 para. 2 BV inclu­des the “Pro­tec­tion of infor­ma­tio­nal self-deter­mi­na­ti­on”, which safe­guards the exer­cise of fun­da­men­tal rights.
  • Dif­fe­rent “Enga­ge­ment tor­ques” may have to be justi­fi­ed indi­vi­du­al­ly, and they must also be rea­sonable when view­ed as a who­le. Infor­ma­tio­nal self-deter­mi­na­ti­on is par­ti­cu­lar­ly impai­red if the data pro­ce­s­sing can lead to a self-rest­ric­tion in the exer­cise of fun­da­men­tal rights (chil­ling effect).
  • The Inten­si­ty The ext­ent to which the­re is an ele­ment of inter­fe­rence results from the inter­play bet­ween the loss of con­trol and the effec­ti­ve­ness of legal reme­dies against it, and from the per­so­nal natu­re of the data pro­ce­s­sing. In indi­vi­du­al cases, it must be exami­ned to what ext­ent a loss of con­trol can be coun­ter­ac­ted by safe­guar­ding mea­su­res and the pro­xi­mi­ty to per­so­na­li­ty by anonymization.
  • The sto­rage of per­so­nal data in the cloud is a Sto­rage “on reser­ve” for the atten­ti­on of U.S. aut­ho­ri­tieswho have access via the CLOUD Act or the Stored Com­mu­ni­ca­ti­ons Act (SCA). Such access would con­sti­tu­te a vio­la­ti­on of the prin­ci­ples of Swiss law, non-trans­pa­rent pro­ce­s­sing and a vio­la­ti­on of pur­po­se limi­ta­ti­on. Due to the num­ber of data sub­jects affec­ted and the loss of con­trol, a serious encroach­ment can gene­ral­ly be assumed.
  • The “Rosen­thal method” appears to be “not yet suf­fi­ci­ent­ly matu­re” for use in the public law sec­tor, but it can pro­vi­de infor­ma­ti­on on the order of magnitu­de of the pro­ba­bi­li­ty of a breach. Moreo­ver, no alter­na­ti­ve with a simi­lar­ly struc­tu­red argu­men­ta­ti­on is appa­rent; the data pro­tec­tion aut­ho­ri­ties also pro­vi­de “litt­le gui­dance on how [the risks asso­cia­ted with cloud use] could be asses­sed in a legal­ly sound manner”.
  • That a Low risk of access seems “plau­si­ble under the given cir­cum­stances”, but this may chan­ge with the incre­a­sing use of M365.
  • Inter­ven­ti­ons gain in inten­si­ty when the Depen­dence on the Office pro­ducts taken into account. The public admi­ni­stra­ti­on has no alter­na­ti­ves for the fur­ther deve­lo­p­ment of the­se products.
  • The legal bases in Zurich allow the pro­ce­s­sing of spe­cial per­so­nal data, but they are not suf­fi­ci­ent for the spe­ci­fic inter­ven­ti­on moments of out­sour­cing to the cloud of a US company.
  • Like can­to­nal aut­ho­ri­ties, the experts recom­mend sto­ring at least sen­si­ti­ve per­so­nal data in the cloud. encryptwhich could enable con­sti­tu­tio­nal­ly com­pli­ant use of M365 if the encryp­ti­on is also effec­ti­ve against access by the provider.

It fol­lows from this:

In view of the inter­ven­ti­ons descri­bed and their inten­si­ty, as well as the ina­de­qua­te legal basis for the assump­ti­on of the cor­re­spon­ding risks by public bodies in the Can­ton of Zurich, the fol­lo­wing must be taken into account at the pre­sent time a wai­ver of cer­tain forms of pro­ce­s­sing of spe­cial per­so­nal data (= serious inter­fe­rence with infor­ma­tio­nal self-deter­mi­na­ti­on) by means of M365 is recom­men­ded. This applies to all forms of pro­ce­s­sing that invol­ve sto­ring data in the Micro­soft cloud. For the fore­seeable future, a mil­der opti­on here is to ope­ra­te such appli­ca­ti­ons in our own data cen­ters and only update them via the cloud ser­vice. Alter­na­tively, tra­di­tio­nal out­sour­cing to data cen­ters of a third par­ty that is not sub­ject to U.S. law can be con­side­red. This assess­ment must be con­stant­ly review­ed in the light of future legal and tech­ni­cal developments.

Cri­ti­cal comments

Argu­ments vs. terminology

The report gives rise to cri­ti­cism on the fol­lo­wing points. First of all, the sub­stan­ti­ve argu­ments must be distin­gu­is­hed from the ter­mi­no­lo­gy, which is not easy. Howe­ver, the ter­mi­no­lo­gi­cal natu­re of the comm­ents on Right to infor­ma­tio­nal self-deter­mi­na­ti­on. This right is a genui­ne right hard­ly justi­fi­ed:

  • The aut­hors start from the cen­sus ruling of the Ger­man Fede­ral Con­sti­tu­tio­nal Court, which reco­gnized such a right. In Switz­er­land, the Fede­ral Court has repea­ted­ly used the con­cept of the right to infor­ma­tio­nal self-deter­mi­na­ti­on. The aut­hors trace this deve­lo­p­ment, as well as the legislator’s refe­ren­ces to this concept.
  • Howe­ver, they con­ce­de that the con­tent of this right can only be deter­mi­ned by inter­pre­ting Artic­le 13 of the Fede­ral Con­sti­tu­ti­on and the rele­vant sta­tu­to­ry law. It fol­lows that infor­ma­tio­nal self-deter­mi­na­ti­on is essen­ti­al­ly only a con­cept that has no nor­ma­ti­ve force; it is the­r­e­fo­re mere­ly a con­cep­tu­al bracket for con­sti­tu­tio­nal and sta­tu­to­ry law to be inter­pre­ted accor­ding to the usu­al rules.
  • The que­sti­on is the­r­e­fo­re whe­ther the term “infor­ma­tio­nal self-deter­mi­na­ti­on” is real­ly appro­pria­te – a dubio­us but pro­ba­b­ly seman­tic que­sti­on; howe­ver, it is of litt­le use as a source of inter­pre­ta­ti­on. The aut­hors hard­ly deri­ve any con­cre­te con­clu­si­ons from this term.

Also of a ter­mi­no­lo­gi­cal natu­re are the refe­ren­ces to the various “Inter­vening fac­tors” of data pro­ce­s­sing. The aut­hors argue that various “intru­si­ve ele­ments” of data pro­ce­s­sing (e.g. type of data, pur­po­se of pro­ce­s­sing, type and scope of pro­ce­s­sing, the fur­ther cir­cum­stances of pro­ce­s­sing, etc., i.e. the “means of pro­ce­s­sing”) must each be legi­ti­mi­zed in their own right under Art. 36 Cst. and that the pro­ce­s­sing must also be rea­sonable over­all. This is cor­rect, but it says litt­le: whe­ther the requi­re­ments of Art. 36 FC are applied to indi­vi­du­al aspects of inter­fe­rence or to pro­ce­s­sing as a who­le is irrele­vant; eit­her way, the seve­ri­ty of the inter­fe­rence is decisi­ve, and this can of cour­se result from indi­vi­du­al aspects such as data sto­rage or their inter­ac­tion or both.

Con­tent point: Loss of control

In the actu­al exami­na­ti­on of the inter­ven­ti­on moments iden­ti­fi­ed in con­nec­tion with cloud out­sour­cing and thus the core of the que­sti­on, the aut­hors come to the con­clu­si­on that out­sour­cing is a Legal and de fac­to loss of con­trol is asso­cia­ted with this. It is this loss of con­trol that leads them to the con­clu­si­on that the exi­sting legal basis is not sufficient.

You begin here by sta­ting that the exi­sting expert opi­ni­ons on the sub­ject argue with the low pro­ba­bi­li­ty of access by the aut­ho­ri­ties, and right­ly so, but that they fail to take into account the fact that the­re are alre­a­dy sto­rage in the cloud is an inter­ven­ti­on ele­ment in its own right which should be exami­ned separately.

This is cer­tain­ly con­cep­tual­ly cor­rect, but not new – if the sto­rage can con­sti­tu­te a dan­ger, it must of cour­se be exami­ned accor­ding to the cri­te­ria of Art. 36 BV, whe­ther it is descri­bed as an “ele­ment of inter­fe­rence” or not. Glass and Sche­fer assu­me a risk asso­cia­ted with mere sto­rage in a US cloud, increa­sed risk  essen­ti­al­ly becau­se the US CLOUD Act and the Stored Com­mu­ni­ca­ti­ons Act adapt­ed by it allow access in cir­cum­ven­ti­on of legal assi­stance, in vio­la­ti­on of the prin­ci­ples of trans­pa­ren­cy and pur­po­se limi­ta­ti­on and in vio­la­ti­on of the Cyber­crime Convention.

The­se Inter­pre­ta­ti­on of the CLOUD Act does not app­ly:

  • The the­sis that sto­rage in a cloud of a US pro­vi­der is a Data reten­ti­on for the atten­ti­on of the US aut­ho­ri­ties is based on the con­side­ra­ti­on that access under the SCA vio­la­tes Art. 32 of the Cyber­crime Con­ven­ti­on (CCC). Art. 32 CCC sti­pu­la­tes that aut­ho­ri­ties of a con­trac­ting sta­te may access data in the ter­ri­to­ry of ano­ther con­trac­ting par­ty essen­ti­al­ly only with the con­sent of the per­son who may dis­po­se of this data (e.g. a local ISP; see also BGE 141 IV 108). Howe­ver, Art. 18 CCC sti­pu­la­tes that data abroad may be acce­s­sed by anyo­ne who has pos­ses­si­on or con­trol over this data. This may be Micro­soft USA, but also depends on the spe­ci­fic design of the ser­vices. This cor­re­sponds more or less to the SCA (§ 2713), which the­r­e­fo­re does not vio­la­te basic rules reco­gnized in Switzerland.
  • It is not the case that, in the event of a sur­ren­der order, a pro­vi­der must Never inform cus­to­mers may. This depends on the legal basis of the sur­ren­der order – in the case of a sub­poe­na or a court order, the infor­ma­ti­on is at least not gene­ral­ly exclu­ded (SCA § 2703(b)(1)).
  • It is also incor­rect that a US pro­vi­der can only defend its­elf against access by US aut­ho­ri­ties based on the SCA if a Exe­cu­ti­ve Agree­ment was con­clu­ded. If it is miss­ing, he can­not cla­im that a sur­ren­der would vio­la­te the law of ano­ther sta­te, but the objec­tion remains that a “comi­ty ana­ly­sis” pre­vents the sur­ren­der (§ 2703(h)(2)(B)(ii) and (3)).
  • The aut­ho­ri­ties’ access opti­ons under the SCA are by no means “unli­mi­t­ed“as the aut­hors wri­te. Rather, they pre­sup­po­se that a war­rant, a sub­poe­na or a court order has been issued. This says not­hing about the pre­re­qui­si­tes for the­se instru­ments, but they are not uncon­di­tio­nal and available “inde­fi­ni­te­ly”.

The aut­hors address some of the­se points in the adden­dum. Howe­ver, as they stand by the con­clu­si­ons of the report, the cri­ti­cism remains relevant.

What remains

The argu­men­ta­ti­on of Glass and Sche­fer is fun­da­men­tal­ly con­si­stent in that they empha­si­ze that more serious encroach­ments on fun­da­men­tal rights requi­re an expli­cit legal basis, i.e. that an indi­rect legal basis through the assign­ment of tasks is no lon­ger suf­fi­ci­ent. The­re is not­hing to be said against this. On the merits, howe­ver, they argue stron­gly – not exclu­si­ve­ly – with an inter­pre­ta­ti­on of the Stored Com­mu­ni­ca­ti­ons Act that can­not be shared. Accor­din­gly, their con­clu­si­ons must also be rejected.

Howe­ver, the refe­rence to a cer­tain loss of con­trol and sup­plier depen­den­cy is cor­rect. Whe­ther the­se aspects in them­sel­ves call for a spe­cial legal basis is left open by the expert opi­ni­on becau­se it does not assu­me that only the­se aspects are decisive.

It the­r­e­fo­re remains the case that the risks of access by the aut­ho­ri­ties must be exami­ned and that, abo­ve all, the Pro­ba­bi­li­ty of access – and, of cour­se, the weight of its con­se­quen­ces – are decisi­ve. Here, the aut­hors do not con­sider the Rosen­thal method to be unsui­ta­ble, even if they reco­gnize cer­tain short­co­mings (which, howe­ver, can­not be descri­bed as short­co­mings; rather, they are inher­ent in the method, which does not cla­im to ans­wer all que­sti­ons). As a result, it can be sta­ted that the expert opi­ni­on should have no influence on the assess­ment of out­sour­cing by public bodies.

Notes from David Rosenthal

As noted, David Rosen­thal has also writ­ten detail­ed comm­ents on the expert opi­ni­on, which available here are. Inso­far as they are con­si­stent with the con­tent of the abo­ve comm­ents, they are not repea­ted here. Howe­ver, it also notes the fol­lo­wing points:

  • Swiss law reco­gnizes sur­ren­der orders that are com­pa­ra­ble to the SCA. It is the­r­e­fo­re also wrong to say that the SCA is dif­fi­cult to recon­ci­le with the prin­ci­ples of Swiss law.
  • The “end-to-end” encryp­ti­on for sen­si­ti­ve data recom­men­ded in the report is neither sui­ta­ble nor neces­sa­ry, at least for M365; access by the aut­ho­ri­ties can also be coun­ter­ac­ted with less rest­ric­ti­ve measures.
  • It is not rele­vant whe­ther the­re is an ade­qua­te level of data pro­tec­tion in the USA (which should be the case in the fore­seeable future with the Pri­va­cy Frame­work, even from a Swiss per­spec­ti­ve). The cus­to­mer dis­c­lo­ses data to Micro­soft in Ire­land. Access from the USA is at most excep­tio­nal, and then not neces­s­a­ri­ly to per­so­nal data.
  • It can­not be said that local sto­rage of per­so­nal data is more secu­re than sto­rage in the cloud. Rather, it must be taken into account that with M365, the “mini­mal loss of con­trol” in the area of access by aut­ho­ri­ties is off­set by a “signi­fi­cant­ly hig­her ‘gain in con­trol’ in terms of pro­tec­tion against hackers and other dangers”.
  • The fun­da­men­tal sui­ta­bi­li­ty of the Rosen­thal method is con­firm­ed in the expert opi­ni­on. The pro­ba­bi­li­ty of access by for­eign aut­ho­ri­ties remains decisi­ve. The method can plau­si­bly demon­stra­te this.