Piasecki, S anislaw; Mo osoli, Sophie; Helbe ge , Na ali; Naud s, Lau ens
A icle
AI-gene a ed jou nalism: Do he anspa ency p o isions
in he AI Ac gi e news eade s wha hey hope o ?
In e ne Policy Re iew
P o ided in Coope a ion wi h:
Alexande on Humbold Ins i u e o In e ne and Socie y (HIIG), Be lin
Sugges ed Ci a ion: Piasecki, S anislaw; Mo osoli, Sophie; Helbe ge , Na ali; Naud s, Lau ens (2024) :
AI-gene a ed jou nalism: Do he anspa ency p o isions in he AI Ac gi e news eade s wha hey
hope o ?, In e ne Policy Re iew, ISSN 2197-6775, Alexande on Humbold Ins i u e o In e ne
and Socie y, Be lin, Vol. 13, Iss. 4, pp. 1-28,
h ps://doi.o g/10.14763/2024.4.1810
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/312555
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by/3.0/de/deed.en
Volume 13 |
AI-gene a ed jou nalism: Do he
anspa ency p o isions in he AI Ac gi e
news eade s wha hey hope o ?
S anislaw Piasecki Hi o subashi Uni e si y
Sophie Mo osoli Uni e si y o Ams e dam
Na ali Helbe ge Uni e si y o Ams e dam
Lau ens Naud s Uni e si y o Ams e dam
DOI: h ps://doi.o g/10.14763/2024.4.1810
Published: 23 Oc obe 2024
Recei ed: 5 Feb ua y 2024 Accep ed: 1 July 2024
Funding: This publica ion is pa o he AI, Media & Democ acy Lab (Du ch Resea ch
Council p ojec numbe : NWA.1332.20.009). The au ho s we e also suppo ed by he
VolkswagenS i ung Founda ion unde g an numbe C.2127.0331.
Compe ing In e es s: The au ho has decla ed ha no compe ing in e es s exis ha
ha e in luenced he ex .
Licence: This is an open-access a icle dis ibu ed unde he e ms o he C ea i e
Commons A ibu ion 3.0 License (Ge many) which pe mi s un es ic ed use,
dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly
ci ed. h ps://c ea i ecommons.o g/licenses/by/3.0/de/deed.en
Copy igh emains wi h he au ho (s).
Ci a ion: Piasecki, S. & Mo osoli, S. & Helbe ge , N. & Naud s, L. (2024). AI-gene a ed
jou nalism: Do he anspa ency p o isions in he AI Ac gi e news eade s wha hey
hope o ?. In e ne Policy Re iew, 13(4). h ps://doi.o g/10.14763/2024.4.1810
Keywo ds: T anspa ency, AI Ac , Media, Gene a i e AI, Jou nalism
Abs ac : Issues linked o he inc easing p esence o AI-gene a ed con en in people’s li es, and he
impo ance o being able o e ec i ely na iga e and dis inguish such con en , a e inhe en ly linked
o anspa ency, a no ion ha ou s udy ocuses on by e alua ing A . 50 o he AI Ac . This a icle is
a call o ac ion o ake he in e es s o end use s in o accoun when speci ying AI Ac 's
anspa ency equi emen s. I ocuses on a speci ic use case – media o ganisa ions p oducing ex
wi h he help o gene a i e AI. We a gue ha in i s cu en o m, A . 50 lea es many unce ain ies
and isks doing oo li le o p o ec na u al pe sons om manipula ion o o empowe hem o ake
p o ec i e ac ions. The a icle combines documen al and su ey da a analysis (based on a sample
ep esen a i e o he Du ch popula ion) o p opose conc e e policy and egula o y
ecommenda ions on he ope a ionalisa ion o he AI Ac ’s anspa ency obliga ions. I s main
objec i e is o espond o he ollowing ques ion: how o econcile AI Ac ’s anspa ency p o isions
applicable o digi al news a icles gene a ed by AI wi h news eade s’ pe cep ions o manipula ion
and empowe men ?
Issue 4
In oduc ion
The Me iam-Webs e dic iona y has chosen “au hen ic” as wo d o he yea o
2023, which unde lines he inc easing p esence o AI-gene a ed con en in peo-
ple’s li es, and he impo ance o being able o e ec i ely na iga e and dis inguish
such con en (I alie, 2023). Being able o do so is inhe en ly linked o anspa ency,
a no ion ha ou s udy ocuses on by e alua ing A . 50.1 and A . 50.4 o he AI
Ac (AI Ac , 2024). The egula ion was published on he 12 h o July 2024 in he
O icial Jou nal o he Eu opean Union and is in o ce since he 1s o Augus .
T anspa ency is “one o he co e alues p omo ed by he EU o he de elopmen ,
deploymen , and use o AI sys ems” (Kisele a, 2021). The impo ance o iden i ying
he sou ce o in o ma ion is also con i med by a ious in e na ional ini ia i es such
as he Adobe Con en Ini ia i e, he objec i e o which is o p omo e he “adop ion
o an open indus y s anda d o con en au hen ici y and p o enance” (Adobe,
n.d.). This a icle is a call o ac ion o ake he in e es s o end use s in o accoun
when speci ying AI Ac 's A . 50 anspa ency equi emen s. I ocuses on a speci ic
use case – media o ganisa ions p oducing ex wi h he help o gene a i e AI. The
s udy a gues ha in i s cu en o m, A . 50 s ill lea es many unce ain ies and
isks doing oo li le o p o ec news eade s om manipula ion o o empowe
hem o ake p o ec i e ac ions. Mo eo e , conside ing he sec o ’s pa icula i ies,
including he alue-d i en app oach o jou nalis s (Bas ian e al., 2021), u he
guidance is needed o he media and policymake s.
Be o e he AI Ac , media p o essionals we e unsu e whe he hey should in o m
hei eade s abou he use o AI in news p oduc ion. This will soon become (in
ce ain ci cums ances) subjec o a legal equi emen . This a icle combines docu-
men al and su ey da a analysis (based on a sample ep esen a i e o he Du ch
popula ion) o p opose conc e e policy and egula o y ecommenda ions on he
ope a ionalisa ion o he AI Ac ’s anspa ency obliga ions (which could be includ-
ed in a code o conduc and guidelines as explained in Sec ion 2). The indings and
sugges ions a e g ounded in empi ical e idence and in he expec a ions o news
eade s. The main objec i e is o espond o he ollowing ques ion: how o econ-
cile he AI Ac ’s anspa ency p o isions applicable o digi al news a icles gene -
a ed by AI wi h news eade s’ pe cep ions o manipula ion and empowe men ?
Fi s ly, his s udy explo es he cu en legal landscape: how o in e p e ans-
pa ency p o isions in A . 50 o he AI Ac in ela ion o digi al news a icles? (Sec-
ion 1)
Secondly, his wo k e alua es h ough su ey da a people’s pe cep ions o manipu-
2 In e ne Policy Re iew 13(4) | 2024
la ion and empowe men in he con ex o news a icles ully o pa ly gene a ed
by AI sys ems. In o de o do so, i analyses he ollowing opics: how much ans-
pa ency and agency (as a esul o he o me ) do people wan when eading news
p oduced by AI e sus news p oduced by humans? How do hey pe cei e manipu-
la ion and empowe men ? (Sec ion 2 – “design and sample”, “measu emen ” and
“ esul s” sub-sec ions.)
Thi dly, Sec ion 3 p oposes how o u he speci y he AI Ac wi h ele an obliga-
ions and a o dances based on he documen al and empi ical indings. Wha kind
o egula o y and policy measu es could help in econciling AI Ac ’s anspa ency
p o isions wi h people’s expec a ions ega ding news consump ion? Why is i im-
po an o policy make s o mee hose expec a ions? (Sec ion 3)
Finally, he a icle is b ie ly concluded by summa ising i s main indings and call-
ing o ac ion o u he speci y AI Ac ’s anspa ency equi emen s.
1. The legal landscape ela ed o anspa ency in he
AI Ac and digi al news a icles
P io o he analysis o he empi ical s udy whe e we asked su ey pa icipan s
abou hei anspa ency expec a ions and ollow-up ac ion p e e ences in he con-
ex o news a icles p oduced by AI, i is impo an o unde s and wha he law is
and how media o ganisa ions need o na iga e he new anspa ency equi e-
men s. To wha ex en a e media p o essionals cu en ly obliged o p o ide in o -
ma ion o news eade s and o wha deg ee is disclosu e le o hei own deci-
sion? A . 50 manda es anspa ency obliga ions o AI sys ems ega dless o
whe he hey a e conside ed as high isk o no (Almada & Pe i , 2023). As he me-
dia is no conside ed high isk in he AI Ac , his is he main anspa ency- ela ed
p o ision applicable o his sec o – “by manda ing disclosu e o he a i icial cha -
ac e o he sys em, he AI Ac seeks o close oppo uni ies o impe sona ion and
decep ion, which can be ha m ul e en i he sys em i sel is no used o a high-
isk pu pose” (Busuioc e al., 2023, p. 93). This is con i med in Reci al 70, which
s a es ha “ce ain AI sys ems in ended o in e ac wi h na u al pe sons o o gen-
e a e con en may pose speci ic isks o impe sona ion o decep ion i espec i e o
whe he hey quali y as high- isk o no ”. Acco ding o he i s pa ag aph o A .
50.1 AI Ac :
P o ide s shall ensu e ha AI sys ems in ended o in e ac di ec ly wi h na u al
pe sons a e designed and de eloped in such a way ha he na u al pe sons
3 Piasecki, Mo osoli, Helbe ge , Naud s
conce ned a e in o med ha hey a e in e ac ing wi h an AI sys em, unless his
is ob ious om he poin o iew o a na u al pe son who is easonably well-
in o med, obse an and ci cumspec , aking in o accoun he ci cums ances and
he con ex o use. […].
In e ms o AI-gene a ed ex , he second pa ag aph o A . 50.4 s a es ha :
Deploye s o an AI sys em ha gene a es o manipula es ex which is
published wi h he pu pose o in o ming he public on ma e s o public in e es
shall disclose ha he ex has been a i icially gene a ed o manipula ed. This
obliga ion shall no apply […] whe e he AI-gene a ed con en has unde gone a
p ocess o human e iew o edi o ial con ol and whe e a na u al o legal
pe son holds edi o ial esponsibili y o he publica ion o he con en .
Be o e e lec ing on he con en o hese p o isions, i is wo h cla i ying some o
he e minology used in his s udy and in he AI Ac . Fi s ly, in ou a icle we dis-
cuss p o ide s (such as Open AI o Mic oso ), na u al pe sons, and deploye s
(while ocusing on he obliga ions and challenges o he la e ) in ela ion o A .
50 o he egula ion. Al hough i is s ill used in a ew eci als, in he inal e sion
o A . 50, he e m “use ” (adop ed in all o he p e iously p oposed d a s) has
been eplaced by “deploye ” ( he media o ganisa ion), which is a welcome de el-
opmen . The o me has been c i icised – o example, by he Ada Lo elace Ins i-
u e (Ci ciuma u, 2022) – as i may lead o con usion on whe he he use is ac u-
ally he end use o he deploye .
Secondly, no all AI con en is decep i e. In he con ex o ou wo k, manipula ion
is no conside ed as due o he sys em jus being AI bu a he due o simply no
knowing ha i is AI. People us and ead some jou nalis s mo e han o he s.
Simila ly, hey may us AI (also in ela ion o a pa icula AI p o ide o e ano h-
e ) mo e han humans (o he e e se) and should be able o decide (i hey wan
o decide) wha sou ces hey p e e . As a consequence, no disclosing ha AI gen-
e a ed an a icle o ha i has been w i en by a pa icula jou nalis would be
equally misleading. In addi ion, al hough no always decep i e, AI unc ions di e -
en ly han human jou nalis s, which u he con i ms he impo ance o being in-
o med abou i s use in news p oduc ion.
A . 50.1 imposes in o ma ion obliga ions only on p o ide s (con a ily o he Eu o-
pean Pa liamen ’s e sion, i does no men ion deploye s) and only when he “AI
4 In e ne Policy Re iew 13(4) | 2024
sys ems a e in ended o di ec ly in e ac wi h na u al pe sons”. The goal o his
p o ision is o in o m he “conce ned na u al pe sons” abou hei in e ac ion wi h
AI, ha is, p o ide s need o design AI sys ems in a way ha makes i possible o
do so. On he one hand, his p o ision ce ainly applies o gene a i e AI cha bo s
(such as Cha GPT) when he con en is di ec ly p esen ed o na u al pe sons as a
esul o hei own que ies. Acco ding o he AI Ac , he la e ha e he igh o be
in o med ha such an in e ac ion occu s. On he o he hand, i could be a gued
ha in he media con ex he e is an in e media y ( he media o ganisa ion) be o e
he con en a i es o he na u al pe son (news eade ) and, as a esul , he e is in-
di ec in e ac ion causing A . 50.1 no o apply. Howe e , one could also con end
ha he AI sys em was s ill “in ended” o di ec ly p o ide in o ma ion o he end
use (na u al pe son). Mo eo e , when he con en p o ided by he deploye is he
same as he one o iginally gene a ed by AI, one could asse ha di ec in e ac ion
s ill occu s. No hing was changed excep he place whe e he con en was ac-
cessed ( o example, he media o ganisa ion’s websi e ins ead o he gene a i e AI
p o ide ’s websi e). Which in e p e a ion will p e ail equi es u he cla i ica ion.
I “di ec ” means di ec in e ac ion wi h he o iginally p oduced con en , his
would signi y ha he p o ide s’ obliga ion o in o m also ex ends o si ua ions
whe e deploye s publish any kind o o iginal AI-gene a ed ma e ial, and ha he
p o ide s’ ma ks, such as wa e ma ks, should no be emo ed. I “di ec ” is in e -
p e ed o s ic ly mean ha na u al pe sons mus gene a e he con en hem-
sel es, hen deploye s (including media o ganisa ions) migh be allowed o use he
AI-gene a ed con en wi hou p o ide s’ ma ks in place and no disclose o news
eade s ha hey a e in e ac ing wi h AI (o disclose i di e en ly). A ye open
ques ion is how A . 50.1 and A . 50.4 ela e o each o he in case o a b oad in-
e p e a ion o “di ec ly in e ac ”. Is disclosing he p o ide ’s ma k enough o also
sa is y he A . 50.4’s deploye in o ma ion obliga ion? O would he media o gani-
sa ion need o in o m he news eade in addi ion o he p o ide ’s label?
Finally, when a pe son sees an AI-gene a ed pic u e, ideo o ex published by a
deploye ( o example, a media o ganisa ion), i will ce ainly no be “ob ious om
he poin o iew o a na u al pe son who is easonably well-in o med, obse an
and ci cumspec ” (as equi ed by A . 50.1) ha hey a e in e ac ing wi h AI. I is
inc easingly di icul o all people o dis inguish AI and human con en (unless
pe haps, o example, a pa icula media o ganisa ion is known o always publish-
ing AI ma e ial). This p o ision is an a gumen in a ou o ex ending he applica-
bili y o he in o ma ion obliga ion imposed on p o ide s by A . 50.1, o si ua ions
whe e he o iginal con en is deployed by a hi d pa y. O he wise, he scope o A
50.1 would be qui e na ow. I one in e ac s di ec ly on Bing’s websi e wi h hei
5 Piasecki, Mo osoli, Helbe ge , Naud s
gene a i e AI sys ems, i will be ce ainly “ob ious” o mos people ha hey a e
in e ac ing wi h AI. Howe e , i a pe son sees AI-gene a ed con en on a media o -
ganisa ion’s YouTube channel o , o example, an AI-gene a ed esponse on an in-
e ne b owse ollowing a sea ch que y, i would be much ha de o con end ha
hey a e awa e o in e ac ing wi h AI.
A . 50.4 explici ly men ions deploye s. I s la es e sion does no only conce n de-
ploye s o deep akes (A . 50.4 pa ag aph 1) bu now also deploye s o AI-gene a -
ed ex (A . 50.4 pa ag aph 2). Fo ex ual con en , he media will need o in o m
na u al pe sons ha hey a e in e ac ing wi h AI only unde speci ic condi ions.
Deploye s mus do so i he a icle is w i en “wi h he pu pose o in o ming he
public on ma e s o public in e es ” and i i has no “unde gone a p ocess o hu-
man e iew o edi o ial con ol and a na u al o legal pe son” does no hold “edi o-
ial esponsibili y o he publica ion o he con en ” (A . 50.4). In addi ion, simi-
la ly o A . 50.1, A . 50.4 indica es ha he con en needs o be “a i icially gene -
a ed o manipula ed” o he in o ma ion obliga ion o apply.
The A . 50.4 p o ision leads o many ques ions. Fi s ly, wha does “a i icially gen-
e a ed o manipula ed” signi y in he con ex o A . 50? When is AI used “enough”
o mean ha in o ma ion has been manipula ed? This a icle a gues ha whene -
e AI w i es pa o a news a icle’s subs ance ( ully gene a ed) o is pa aph ased
by a jou nalis (manipula ed), in o ma ion abou his should be p o ided. While
bo h ex s gene a ed by AI and by jou nalis s may con ain mis akes, hei causes
and p e alence di e ge and, as men ioned abo e, he au ho s a e simply no he
same. Fo his eason, news eade s should ha e a choice in e ms o which ype o
au ho hey p e e . To he con a y, i AI is only used o esea ch pu poses, hen in-
o ma ion p o ision should no be necessa y.
Secondly, i is unce ain how o in e p e he “ma e s o public in e es ” condi ion.
A ixed de ini ion o he la e is di icul o p o ide as i di e s depending on he
con ex ( u he legal cla i ica ion is needed). I he c i e ia would be he po en ial
each o he con en , any a icle may become popula and gain people’s in e es ,
especially h ough social media. Always accu a ely p edic ing which ype o news
will do so is no possible. This may esul in impo an p ac ical di icul ies o e -
ec i ely implemen his p o ision. I he c i e ia would be he ype o con en ( o
example, spo esul s e sus poli ical news), hen his di e en ia ion may be pos-
sible. Howe e , i should be based on ele an empi ical esea ch a he han as-
sump ions. Ou su ey da a showed (as will be discussed la e in his s udy) ha
people wan in o ma ion abou he sou ce o news ega dless o whe he he news
is con o e sial o no . Mo eo e , di e en ia ing be ween di e en ypes o opics
6 In e ne Policy Re iew 13(4) | 2024
migh in p ac ice esul in mo e wo k o media o ganisa ions. The “public in e es ”
ac o o a pa icula ype o con en has always been di icul o de e mine and
he e is no one me hod o e alua e i (Caple, 2018, p. 10). Always labelling AI-
gene a ed news could be a mo e e ec i e solu ion (and could lead o mo e us
om na u al pe sons).
Thi dly, assuming he “ma e s o public in e es ” condi ion is sa is ied, o he in-
o ma ion p o ision obliga ion o apply, he con en mus also no unde go “a
p ocess o human e iew o edi o ial con ol”. Wha does his signi y? I a human
ac -checks an a icle w i en by AI, is he disclosu e obliga ion li ed? As a gued
abo e, he decision on whe he o in o m o no should no be aken based on he
con en being decep i e (as he legisla o seems o sugges ) bu a he because
news eade s should ha e he igh o be in o med who p oduced he con en
(e en i a human o he edi o ac -checks i , he sou ce would s ill be AI). As will
be discussed in he nex sec ion, his a icle’s empi ical indings con i m ha he
public expec s his in o ma ion. In addi ion, is no all con en published by media
o ganisa ions o a ce ain ex en unde edi o ial con ol? A b oad in e p e a ion o
his condi ion would lea e only a e y na ow scope o applica ion o he p o i-
sion. Fo his eason, his wo k conside s ha he condi ion o “human e iew o
edi o ial con ol” should be in e p e ed as meaning ha he AI-gene a ed ex
mus be su icien ly ans o med (no jus pa aph ased) by he media p o essional
(as a esul , i would no be “a i icially gene a ed o manipula ed” anymo e). All
ex s w i en on “ma e s o public in e es ” should disclose he use o gene a i e
AI sys ems unless su icien ly ans o med unde he edi o ’s con ol.
Fou hly, he main objec i e o A . 50 is o in o m na u al pe sons abou hei in-
e ac ion wi h AI (Hacke , 2023). This p o ision could be in e p e ed s ic ly, as sig-
ni ying ha simply in o ming abou he exis ence o an in e ac ion is su icien .
Howe e , i could also imply ha addi ional in o ma ion should be disclosed (such
as he name o he AI p o ide ’s company). Mo eo e , anspa ency can be “an im-
po an means o imp o e p ocedu al igh s” i such igh s we e o be gi en o
news eade s (Va ošanec, 2022, p. 95). As men ioned in Reci al 14 o he AI Ac :
T anspa ency means ha AI sys ems a e de eloped and used in a way ha
allows app op ia e aceabili y and explainabili y, while making humans awa e
ha hey communica e o in e ac wi h an AI sys em, as well as duly in o ming
deploye s o he capabili ies and limi a ions o ha AI sys em and a ec ed
pe sons abou hei igh s (emphasis added).
7 Piasecki, Mo osoli, Helbe ge , Naud s
The AI Ac explici ly s a es ha “a ec ed pe sons” should no only be in o med
abou he ac ha hey “communica e o in e ac ” wi h AI bu also abou igh s
ha hey can exe cise as a di ec consequence o in o ma ion p o ision. In o ma-
ion is gi en o empowe people and p o ide hem wi h mo e con ol (no only o
in o m) (The Ams e dam Pape , 2024). A alid ques ion is he e o e o ask wha in-
o ma ion abou he in e ac ion should be p o ided o he na u al pe son when
he la e eads news a icles gene a ed by AI ( o educe manipula ion)? Wha kind
o igh s should news eade s possess in his con ex ? How o e ec i ely empowe
news eade s o exe cise hose igh s?
The e is s ill much o be discussed in e ms o how he AI Ac should be imple-
men ed and wha kind o policy and implemen a ion measu es need o be adop ed
o do so. These ques ions will be explo ed in Sec ion 2. This is especially c ucial
o he media sec o whe e alues and building na u al pe sons’ us in an e hical
manne a e an essen ial pa o jou nalis ic codes o conduc and wo k p ocesses –
“ anspa ency in he iew o he law is no a goal in i sel , bu a means ha is
needed o p omo e a ange o e y di e en alues” (Gye na e al., 2023).
2. Di e ences be ween human and AI-gene a ed news
– Empi ical analysis o news eade s’ expec a ions and
hei econcilia ion wi h he AI Ac
Ha ing iden i ied ha anspa ency plays an impo an ole in he AI Ac o he
media sec o , and hus in luences how con en will be p esen ed o na u al pe -
sons in he nea u u e, i is essen ial o unco e how people eac o anspa ency
cues in he con ex o news con en . Cen al ques ions connec ed o his a ise:
would news eade s eel manipula ed i hey we e no in o med abou he ac ha
an a icle was p oduced by AI? Wha do people do wi h news con en once hey
ha e been con on ed wi h a anspa ency cue? How would hei in e es in being
able o exe cise con ol o e news con en di e o human and AI w i en con-
en ? Fo ins ance, would hey like o ha e addi ional in o ma ion abou news p o-
duc ion o ha e he op ion o il e news a icles om a ce ain sou ce? The ol-
lowing sec ion empi ically ackles hese challenges. As men ioned by Ha esamu-
d am, “use -cen ed esea ch on AI anspa ency emains limi ed” (Ha esamud am
e al., 2023, p. 99). This con ibu ion should be seen as only one elemen o a la g-
e and needed discussion on how o ope a ionalise and speci y AI Ac ’s A . 50 in
he media sec o . The empowe men - ela ed issues we ackled a e no exhaus i e
and he e migh be o he ele an opics equi ing empi ical esea ch.
8 In e ne Policy Re iew 13(4) | 2024
ga ding people’s ollow up ac ions when hey ead news headlines, as well as hei
pe cep ions o manipula ion and empowe men . Mo ing o wa d, we pu he ind-
ings in o a bigge pic u e.
Based on he esul s o his s udy, we iden i ied he need o anspa ency o coun-
e ac he eeling o manipula ion. In o he wo ds: news eade s wan o know
whe he an a icle has been w i en by AI o no , i espec i e o whe he ha con-
en is liable o in luence he b oade public opinion and i espec i e o hei be-
lie s o whe he AI-gene a ed con en is mo e o less accu a e, us wo hy, alue-
o ien ed, o has hei bes in e es s in mind. Inso a , A . 50 AI Ac is impo an
om he pe spec i e o news eade s. Howe e , ou esea ch also shows he cu -
en limi a ion in A . 50.4, ha exemp s con en ha has unde gone a p ocess o
human e iew o edi o ial con ol and whe e a na u al o legal pe son holds edi o-
ial esponsibili y o he publica ion, is con a y o he in e es s o he audience.
People would eel manipula ed i no in o med abou he syn he ic o igin o he
con en , e en i i has unde gone edi o ial e iew (unless, as sugges ed in Sec ion
1, he edi o ial e iew condi ion would be in e p e ed as meaning ha he ex
needs o be modi ied o he ex en ha i is no possible o conside i s con en as
“a i icially gene a ed o manipula ed” anymo e). Mo eo e , i no in o ma ion is
p o ided, people would no be able o exe cise any o he ollow-up ac ions in e-
la ion o he empowe men beha iou s measu ed h ough his s udy’s su ey da a
(as men ioned in he p e ious sec ion, an impo an numbe o news eade s
would wan o do so).
Impo an ly, ou esea ch also inds ha simply in o ming people abou he ac
ha a ex has been au oma ically gene a ed o manipula ed by AI gi es hem li -
le cues abou how o in e p e and assess he a icle. This may also explain why
we saw no signi ican di e ences in hei in ended ollow-up beha iou . Indeed,
ou esul s sugges ha being in o med abou he ac ha a headline was gene -
a ed by AI o w i en by a human did no signi ican ly a ec hei willingness o
sha e, con inue eading o e en pay o he con en . Con a y o ou indings, Al ay
and Gila di (2023) obse ed ha people we e less willing o sha e news headlines
labelled as AI-gene a ed. The au ho s explained his ou come wi h a dec ease in
accu acy pe cep ion. AI-gene a ed headlines we e pe cei ed less accu a e and hus
news eade s we e less willing o sha e hem. Fu he mo e, he AI-gene a ed
anspa ency labels did no educe he us in news o jou nalis s and he au ho s
could no ind any signi ican g oup di e ences ega ding his ela ionship (Al ay
& Gila di, 2023).
These inconclusi e indings aise a mo e undamen al ques ion abou he ans-
15 Piasecki, Mo osoli, Helbe ge , Naud s
pa ency obliga ions in A . 50.4 pa ag aph 2 o he AI Ac : wha exac ly is and can
be he goal o his p o ision? I he goal is o empowe people o make in o med
decisions on how us wo hy o quali a i e syn he ic con en is, A . 50 o he AI
Ac is likely o ail ha goal. Fo he same eason, he anspa ency p o isions a e
unlikely o sol e he p oblem o popula misconcep ions, misleading imagina ies,
and olk s o ies abou AI (Jasno , 2015; Ca e & Dihal, 2019). Simply in o ming na -
u al pe sons ha con en has been AI gene a ed o manipula ed does no con ey
enough in o ma ion o decide whe he syn he ic con en is us wo hy and lea es
i o news eade s o d aw hei own conclusions based on wha e e hei ideas o
imagina ies o AI a e, o hei le el o knowledge. An in e es ing ques ion o u -
he esea ch could be whe he simple labels can no e en ein o ce pe sis en
imagina ies o machine au onomy and AI aking o e e e la ge pa s o socie y.
Fo he same eason, he in o ma ion obliga ion in A . 50 will no be pa icula ly
use ul as a ool o igh disin o ma ion (as hin ed a in Reci al 70 AI Ac ). In ha e-
ga d, his second ex -based condi ion o A . 50.4 di e s om he deep ake condi-
ion in he i s pa o he pa ag aph. Deep akes a e de ined as “image, audio o
ideo con en ha app eciably esembles exis ing pe sons, places o e en s and
would alsely appea o a pe son o be au hen ic” (Reci al 70b). He e he goal o
anspa ency is o wa n end use s ha he con en mimics eal pe sons o places
bu is no au hen ic. In con as , he second pa ag aph o A . 50.4 AI Ac only e-
qui es ha ex mus be in ended o in o m he public, which only says some hing
abou he unc ion o he ex , no whe he he e en s o ac s i desc ibes a e au-
hen ic o inau hen ic. The objec i e o A . 50.4 emains unclea .
Is he goal o A . 50.4 pa ag aph 2 AI Ac hen o empowe and enable na u al
pe sons o exe cise hei igh s agains syn he ically gene a ed o manipula ed
con en ? Again, he simple ac ha a piece o ex is syn he ic does no gi e
enough in o ma ion o enable na u al pe sons o assess i hei igh s a e a ec ed,
wha hose igh s a e, and how hey could be exe cised. Simply p oducing syn he -
ic con en is no agains he law, nei he does he AI Ac in i s cu en o m gi e
people any igh s o in e ene. A . 85 AI Ac o esees a igh o lodge a complain
wi h a ma ke su eillance au ho i y bu only i a use has easons o belie e ha
he p o isions o he egula ion, such as he anspa ency obliga ions, ha e been
in inged. Fu he below we will discuss wha igh s o en i lemen s news eade s
would like o see.
This lea es digni a ian a gumen s as he main goal ha A . 50.4 pa ag aph 2 may
se e, namely ha na u al pe sons ha e a mo al igh o be in o med i hey a e
subjec ed o AI-gene a ed o manipula ed con en . The ques ion is hen: why
16 In e ne Policy Re iew 13(4) | 2024
would he anspa ency obliga ions only apply o ex ual con en ha is in ended
o in o m on public ma e s, o is excluded o con en unde edi o ial con ol? In
conclusion: he anspa ency p o isions in A . 50.1 and A . 50.4 o he AI Ac a e
impo an and necessa y, also om he pe spec i e o na u al pe sons, bu simply
no ye su icien ly hough h ough.
Wo ying is he inding ha people mis us 5 news sou ces in gene al – ega dless
o whe he an a icle was w i en by an AI o a human jou nalis . The media is
gene ally belie ed o ha e an impo an ole in p o iding us wo hy and quali y
in o ma ion as an an ido e o human o au oma ed disin o ma ion. The ac ha a
signi ican po ion o esponden s we e no con inced ha media con en is us -
wo hy, will gi e hem he in o ma ion hey need, and ha e hei in e es in mind
mus be ood o hough o he media sec o .
Nex o wan ing o be in o med abou he ac ha a piece o con en has been
gene a ed by, o manipula ed wi h AI, ou s udy also shows ha people wan mo e
han anspa ency: hey wan a choice and be able o exe cise a ce ain le el o
agency. By agency, his a icle means “ he exe cise o mani es a ion o one’s capac-
i y o ake ac ions, o ‘do hings’” (And ada e al., 2023, p. 1327), which we also
iew as a componen o news eade s’ empowe men . The s udy by And ada and
colleagues aligns wi h ou indings ha simply in o ming end use s does no in-
c ease hei agency and ha ele an mechanisms migh need o be implemen ed
o do so: “while we assessed only whe he being exposed o a no ice inc eased in-
di iduals’ agency and no he easons behind, he lack o agency could be ex-
plained by he ac ha he pa icipan s could no in luence he in e ac ion wi h AI
(e.g. op ing ou o AI powe ed in e ac ions)” (De And ade e al., 2023, p. 23).
T anspa ency holds bo h in insic and ins umen al alue. To ealise i s ins umen-
al unc ion howe e , we a gue, disclosu e obliga ions should in o m and di ec
news eade s on how o egain and exe cise agency o e hei news en i onmen .
The in o ma ion ha should be p o ided hen, depends upon he o ms o agency
we wish o secu e ( h ough he law), as well as he isks we wan o gua d and em-
powe people agains . As digi al landscapes ha e become cha ac e ised by g ea
asymme ies in powe and knowledge o e echnology (Helbe ge e al., 2021,
2022), a ich body o li e a u e exis s on ( he limi a ions o ) ci izen empowe men ,
including he ole anspa ency, as well as echnical solu ions and he law, can
5. In he con ex o his s udy, we conside and measu e mis us based on he addi ional analysis e-
ga ding us in he in o ma ion and us in sou ce (see Appendix C), he low willingness o pay o
news con en , and he s ong eeling o manipula ion ac oss he s imuli. We a e awa e ha di e -
en measu emen s o mis us exis ; howe e , we belie e ha we cap u e he concep ex ensi ely.
17 Piasecki, Mo osoli, Helbe ge , Naud s
play he ein (see among o he s: Mickli z e al., 2017; an Ooijen & V abec, 2019;
Jablonowska & Palka, 2019; Felzmann e al., 2019, 2020; Lippi e al., 2020; El Ali
e al., 2024). To u he subs an ia e anspa ency’s agency- ela ed unc ion, we
asked su ey pa icipan s wha o ms o agency hey would ind mos desi able.
Th ee ac ions s ood ou : being able o il e news ha has been w i en by an AI o
a human; being able o epo an a icle, espec i ely o complain o a news o gan-
isa ion; and being able o lag biases ha hey see in news p oduced by a pa icu-
la sou ce. Fil e ing a icles om a speci ic sou ce could be seen as he choice o
omi a icles and cu a e people’s news eeds so ha hey do no display a icles
om his sou ce, and ha pe sons ha e he powe o pe sonalise hei news web-
si e o applica ion. The o he wo in e en ions a e mo e a ge ed a di ec ly in e -
ac ing wi h he news o ganisa ion, whe eby i is no clea whe he o con es he
use o AI speci ically, o mo e gene ally o exp ess a wish o mo e in e ac ion and
esponsi eness on he side o he media. The abili y o complain ecalls ea lie de-
mands om he Eu opean Pa liamen o gi e people he possibili y o objec
agains he applica ion o AI sys ems. Wi hin he la e , A . 50.1 second pa ag aph
(o he EP’s o iginal e sion) s a ed ha in o ma ion “shall also include which unc-
ions a e AI enabled, i he e is human o e sigh , and who is esponsible o he
decision-making p ocess, as well as he exis ing igh s and p ocesses” ha allow
o objec agains he applica ion o AI sys ems, “ o seek judicial ed ess agains de-
cisions aken by o ha m caused by AI sys ems”, including he igh o seek an ex-
plana ion.
A guably, o gi e na u al pe sons su icien cues o be able o assess he quali y
and us wo hiness o an AI-gene a ed o manipula ed ex , hey would need ad-
di ional in o ma ion, such as he capabili ies o limi a ions o he a i icial ex
gene a o , o whe he he ex has been subjec o edi o ial con ol.6 P o iding de-
ailed in o ma ion abou whe e and how AI has been applied in he news p oduc-
ion and who is esponsible o he decision-making p ocess can help end use s o
be e unde s and he abili ies bu also he limi s o AI. Fo ins ance, in o ming
eade s i a i icial in elligence has been used o c ea e he headline o he ease
o an a icle (whe e) and i he media company uses hei own AI sys em o i hey
ely on an exis ing one (how). E en hough mo e ex ensi e in o ma ion obliga ions
we e no on he op o he lis o he mos p e e ed empowe men ac ions, s ill al-
mos a hi d o ou esponden s wan ed o ha e mo e in o ma ion han a simple
6. See o example Open AI’s Publica ion and Sha ing policy, sugges ing o include a s a emen along
he lines o : “ he au ho gene a ed his ex in pa wi h GPT-3, OpenAI’s la ge-scale language-gen-
e a ion model. Upon gene a ing d a language, he au ho e iewed, edi ed, and e ised he lan-
guage o hei own liking and akes ul ima e esponsibili y o he con en o his publica ion”
(OpenAI, 2022).
18 In e ne Policy Re iew 13(4) | 2024
cue whe he o no a ex has been p oduced by AI.
Implica ions
The indings om his s udy a e ele an o bo h policymake s and he media. Fo
media o ganisa ions hey sugges ha being anspa en o news eade s abou
he ac ha a piece o con en has been AI gene a ed o manipula ed is pa a-
moun , e en i he AI Ac migh exemp he (edi o ial) media om ha obliga ion.
Telling people ha con en has been AI gene a ed can ha e consequences: some
migh s op eading he con en , il e i ou (i o e ed he op ion), o be less will-
ing o pay o i , bu he consequences o no elling could e en be wo se and
h ea en he al eady agile us ela ionship: people would eel manipula ed. To
some ex en , he indings o his s udy a e also encou aging: in o ming end use s
ha a piece o con en has been AI gene a ed seems o ha e no di ec e ec on
he us wo hiness o pe cep ions o quali y o he con en i sel , o hei willing-
ness o con inue eading o sha ing he con en . The s udy also con i ms ea lie
indings ha na u al pe sons would alue mo e choice and abili y o exe cise
agency and oice (Monze e al., 2020). While he e can be clea economic, s a e-
gic, p agma ic, and o ganisa ional easons agains o e ing people mo e choice,
doing so could also p esen media o ganisa ions wi h an oppo uni y o mend
hei ela ionship wi h he audience and p o ile hemsel es in a ole as s ewa ds o
mo al compass in an inc easingly complica ed and hos ile digi al en i onmen .
Lewis and colleagues, o example, in oduced he concep o ecip ocal jou nalism
and a gued ha “by mo e eadily acknowledging and ecip oca ing he inpu o au-
diences, and by os e ing spaces o audiences o ecip oca e wi h each o he , jou -
nalis s can begin o ul il hei no ma i e pu pose as s ewa ds o he communi ies
hey se e” (Lewis e al., 2014, pp. 236-237). Pu di e en ly, media o ganisa ions
do ha e an oppo uni y he e o do mo e han playing by he ules. Ins ead, hey
can use anspa ency and explainabili y as a means o di e en ia e hemsel es
om la ge echnology co po a ions, whose main in e es is o ge as many people
as possible “hooked” on hei se ices, and explain how hey make su e end use s
can us hei con en .
Fo policymake s, his a icle’s indings a e ood o hough oo. F om he poin o
iew o na u al pe sons, he ac ual e ec i eness o he cu en anspa ency oblig-
a ion is ques ionable. As sugges ed ea lie al eady by he Eu opean Pa liamen ,
anspa ency wi hou agency is no much mo e han a label. And in he case o A .
50 AI Ac in i s p esen o m, ha label does no con ey enough meaning ul in o -
ma ion.
19 Piasecki, Mo osoli, Helbe ge , Naud s
The AI Ac does no discuss delega ed ac s o implemen ing ac s in ela ion o Ti-
le IV on he “T anspa ency Obliga ions o P o ide s and Deploye s o Ce ain AI
Sys ems”. As a esul , addi ional manda o y equi emen s speci ying A . 50 canno
be adop ed based on he egula ion’s p o isions. Howe e , A . 95 men ions he
possibili y o de eloping codes o conduc :
The Commission and he Membe S a es shall acili a e he d awing up o codes
o conduc conce ning he olun a y applica ion, including by deploye s, o
speci ic equi emen s o all AI sys ems, on he basis o clea objec i es and key
pe o mance indica o s o measu e he achie emen o hose objec i es.
In addi ion, A . 96 AI Ac indica es ha he Commission shall de elop guidelines
on he p ac ical implemen a ion o he egula ion, including on “ he p ac ical im-
plemen a ion o anspa ency obliga ions laid down in A . 50”. To do so, he Com-
mission will need mo e cla i y o wha he goal o A . 50 AI Ac is: o in o m o
he sake o in o ming (digni a ian a gumen s), o wa n, o empowe ? Depending on
wha he answe is, in o ming na u al pe sons will need o ake di e en o ms
and should be accompanied by di e en empowe men measu es – he abili y o
lag con en , o il e con en ou , o complain, bu also an obliga ion on he side o
de elope s and deploye s o explain wha he ac ual implica ions a e i a piece o
con en has been AI gene a ed o manipula ed and why i should be us ed.
Di ec ions o u u e esea ch
Fu u e legal-empi ical esea ch can help o design AI anspa ency labels ha e-
spond o he in o ma ion needs o an audience wi h di e en le els o AI li e acy
and ha con ey he in o ma ion ha is ele an o empowe news eade s. O he
po en ial a enues o u u e esea ch ha ou s udy aises conce n he possible
need o di e en ia e be ween AI anspa ency labels o he news media and so-
cial media, he way anspa ency labels a ec ac ual use beha iou , and could po-
en ially esul in less desi able side-e ec s, like e-en o cing socio- echnical
imagina ies and olk s o ies a ound AI.
Ano he issue ha equi es u he esea ch (and ha should be conside ed in
guidelines o a code o conduc ) is how ele an in o ma ion should be communi-
ca ed. A discussion eme ged on wa e ma king as a po en ial ool o achie e com-
pliance wi h he A . 50.1 in o ma ion obliga ions o p o ide s – “implemen a ion
o hese obliga ions [ anspa ency equi emen s o he AI Ac ] will likely equi e
use o wa e ma king echniques” (Madiega, 2023). In e ms o bo h p o ide and
20 In e ne Policy Re iew 13(4) | 2024
deploye - ela ed anspa ency, A . 50.5 s a es ha “ he in o ma ion e e ed o in
pa ag aphs 1 o 4 shall be p o ided o he na u al pe sons conce ned in a clea
and dis inguishable manne a he la es a he ime o he i s in e ac ion o ex-
posu e”. I also adds ha “ he in o ma ion shall espec he applicable accessibili y
equi emen s”. This pape a gues ha o comply wi h hese equi emen s, inspi a-
ion could be d awn om o he ields, such as da a p o ec ion law, also by unde -
s anding he la e ’s limi a ions (P i i e al., 2023). Indeed, anspa ency- ela ed
condi ions – in ela ion o in o ma ion p o ision – ha e been widely deba ed in
he con ex o he Gene al Da a P o ec ion Regula ion (GDPR) (Busuioc e al., 2023;
Naud s e al., 2022). Fo example, in elligibili y is one o he GDPR anspa ency
obliga ions (and was also a equi emen in he Eu opean Pa liamen ’s e sion o
A . 50). As he Cou o Jus ice o he Eu opean Union s a ed in he Kásle case,
he in elligibili y and plain language condi ions “canno … be educed me ely o
hei being o mally and g amma ically in elligible”, bu a he need o be comp e-
hended in a “b oad sense” aking in o accoun an “a e age consume , who is ea-
sonably well in o med and easonably obse an and ci cumspec ” (Case C-26/13).
This is in-line wi h he “ easonably well-in o med, obse an and ci cumspec ” e -
minology used in A . 50 o he AI Ac . As a esul , a jou nalis o a media o ganisa-
ion should i s de ine hei a ge audience and es ablish he a e age audience
membe 's le el o unde s anding. Howe e , in p ac ice, i would be di icul o as-
ce ain who is accessing a pa icula websi e. Fo his eason, his a icle a gues o
always assume ha a ulne able pe son (such as a child o ulne able adul ) could
in e ac wi h a news a icle and o adap communica ion mechanisms by de aul o
such ci cums ances (Piasecki & Chen, 2022). This would make in o ma ion clea e
o e e yone. This is also in-line wi h A . 50.5, which men ions “applicable acces-
sibili y equi emen s” and EU egula ion mo e b oadly. Fo example, he GDPR
anspa ency p inciple manda es o ganisa ions o adop special measu es when
hey p o ide in o ma ion o ulne able people (Piasecki, 2023, p. 13). I is beyond
he scope o his esea ch o analyse communica ion mechanisms in de ail, and
hey would be dependen on he na u e o he in o ma ion p o ided. This a icle
simply wan s o unde line he impo ance o aking his (o en o e looked) aspec
o in o ma ion p o ision in o accoun .
Finally, his a icle does no discuss he ela ionship be ween he AI Ac ’s ans-
pa ency p o isions and consume law. Whe he he new AI egula ion comple-
men s o unde mines he la e , as well as mo e gene ally how o in e p e A . 50
in ligh o cu en consume law p o isions, equi es u he s udy.
21 Piasecki, Mo osoli, Helbe ge , Naud s
Conclusion
T anspa ency is no panacea, also no in enabling end use s o e alua e he sa e y
and us wo hiness o AI-gene a ed con en . Yes, people do wan o know whe he
hey a e exposed o human-w i en o syn he ic con en , and hey would eel ma-
nipula ed i ha in o ma ion was wi hheld om hem. This alone is a s ong ea-
son o jus i y he inclusion o he anspa ency obliga ions in A . 50. And ye , i is
also impo an o ealise ha in i s cu en o m, he anspa ency obliga ion o
AI-gene a ed ex is oo na ow o li e up o he expec a ions o he audience in
being in o med and oo limi ed o con ey much mo e cues han ha a con en has
been a i icially gene a ed. In en yea s om now, syn he ic con en may e y well
cons i u e a la ge sha e o con en on he in e ne – wha hen will be he added
alue o he p o ision? Ins ead, he anspa ency obliga ions in A . 50 should be
he beginning o a con e sa ion on how o make anspa ency meaning ul o he
audience.
Re e ences
Adobe. (n.d.). Con en au hen ici y ini ia i e. Con en Au hen ici y. h ps://con en au hen ici y.o g
Almada, M., & Pe i , N. (2023). The EU AI Ac : A medley o p oduc sa e y and undamen al igh s?
(Wo king Pape No. RSC 2023/59). Robe Schuman Cen e o Ad anced S udies. h ps://cadmus.eu
i.eu/bi s eam/handle/1814/75982/RSC_WP_2023_59.pd ?sequence=1&isAllowed=y
Al ay, S., & Gila di, F. (2023). People a e skep ical o headlines labeled as AI-gene a ed, e en i ue o
human-made, because hey assume ull AI au oma ion. PsyA Xi . h ps://doi.o g/10.31234/os .io/83k9
And ada, G., Clowes, R. W., & Sma , P. R. (2023). Va ie ies o anspa ency: Explo ing agency wi hin
AI sys ems. AI & Socie y, 38(4), 1321–1331. h ps://doi.o g/10.1007/s00146-021-01326-6
Bas ian, M., Helbe ge , N., & Makho ykh, M. (2021). Sa egua ding he jou nalis ic DNA: A i udes
owa ds he ole o p o essional alues in algo i hmic news ecommende designs. Digi al
Jou nalism, 9(6), 835–863. h ps://doi.o g/10.1080/21670811.2021.1912622
Busuioc, M., Cu in, D., & Almada, M. (2023). Reclaiming anspa ency: Con es ing he logics o
sec ecy wi hin he AI Ac . Eu opean Law Open, 2(1), 79–105. h ps://doi.o g/10.1017/elo.2022.47
Case C-26/13. (n.d.). Judgmen o he Cou (Fou h Chambe ), 30 Ap il 2014: Á pád Kásle and Hajnalka
Kásle né Rábai OTP Jelzálogbank Z . The Cou o Jus ice o he Eu opean Union. h ps://eu -lex.eu
opa.eu/legal-con en /EN/TXT/?u i=CELEX:62013CJ0026
Ca e, S., & Dihal, K. (2019). Hopes and ea s o in elligen machines in ic ion and eali y. Na u e
Machine In elligence, 1, 74–78. h ps://doi.o g/10.1038/s42256-019-0020-9
Ci ciuma u, A. (2022). People, isk and he unique equi emen s o AI: 18 ecommenda ions o
s eng hen he EU AI Ac [Policy b ie ing]. Ada Lo elace Ins i u e. h ps://www.adalo elaceins i u e.o
g/policy-b ie ing/eu-ai-ac /
22 In e ne Policy Re iew 13(4) | 2024
De And ade, N., Galindo, L., Za a, A., Heal, J., & Rom, S. (2023). Towa ds in o med AI in e ac ions:
Assessing he impac o no i ica ion s yles on use awa eness and us (Open Loop A i icial
In elligence Ac : A Policy P o o yping Expe imen Ope a ionalizing he Requi emen s o AI Sys ems
– Pa III, pp. 1–27). Open Loop. h ps://openloop.o g/wp-con en /uploads/2023/06/AI_Ac _Towa d
s_In o med_AI_In e ac ions.pd
El Ali, A., Venka aj, K. P., Mo osoli, S., Naud s, L., Helbe ge , N., & Cesa , P. (2024). T anspa en AI
disclosu e obliga ions: Who, wha , when, whe e, why, how. Ex ended Abs ac s o he CHI Con e ence
on Human Fac o s in Compu ing Sys ems, 1–11. h ps://doi.o g/10.1145/3613905.3650750
EY. (2023). Poli ical ag eemen eached on he EU A i icial In elligence Ac (pp. 1–15) [Repo ]. h p
s://asse s.ey.com/con en /dam/ey-si es/ey-com/en_gl/ opics/ai/ey-eu-ai-ac -poli ical-ag eemen -o
e iew-10-decembe -2023.pd
Felzmann, H., Fosch-Villa onga, E., Lu z, C., & Tamò-La ieux, A. (2019). T anspa ency you can us :
T anspa ency equi emen s o a i icial in elligence be ween legal no ms and con ex ual conce ns.
Big Da a & Socie y, 6(1). h ps://doi.o g/10.1177/2053951719860542
Felzmann, H., Fosch-Villa onga, E., Lu z, C., & Tamò-La ieux, A. (2020). Towa ds anspa ency by
design o a i icial in elligence. Science and Enginee ing E hics, 26, 3333–3361. h ps://doi.o g/10.1
007/s11948-020-00276-4
Gye na , B., Fe guson, N., & Scha e , B. (2023). B idging he anspa ency gap: Wha can explainable
AI lea n om he AI Ac ? (Ve sion 5). a Xi . h ps://doi.o g/10.48550/ARXIV.2302.10766
Hacke , P. (2023). AI egula ion in Eu ope: F om he AI Ac o u u e egula o y challenges. a Xi . h p
s://doi.o g/10.48550/a Xi .2310.04072
Ha esamud am, K., La sson, S., & Hein z, F. (2023). Th ee le els o AI anspa ency. Compu e , 56(2),
93–100. h ps://doi.o g/10.1109/MC.2022.3213181
Helbe ge , N., Lynskey, O., Mickli z, H.-W., Ro , P., Sax, M., & S ycha z, J. (2021). EU consume
p o ec ion 2.0: S uc u al asymme ies in digi al consume ma ke s [Repo ]. BEUC The Eu opean
Consume O ganisa ion. h ps://www.beuc.eu/si es/de aul / iles/publica ions/beuc-x-2021-018_e
u_consume _p o ec ion_2.0.pd
Helbe ge , N., Sax, M., S ycha z, J., & Mickli z, H.-W. (2022). Choice a chi ec u es in he digi al
economy: Towa ds a new unde s anding o digi al ulne abili y. Jou nal o Consume Policy, 45(2),
175–200. h ps://doi.o g/10.1007/s10603-021-09500-5
I alie, L. (2023, No embe 27). Wha ’s Me iam-Webs e ’s wo d o he yea o 2023? Hin : Be ue o
you sel . AP News. h ps://apnews.com/a icle/me iam-webs e -wo d-o -yea -2023-a9 ea610cb32e
d913bc15533acab71cc
Jabłonowska, A., & Pałka, P. (2019). EU consume law and a i icial in elligence. In L. De Almeida, M.
C. Gami o, M. Du o ic, & K. P. Pu nhagen (Eds.), The ans o ma ion o economic law: Essays in honou
o Hans-W. Mickli z (pp. 91–112). Ha Publishing. h ps://doi.o g/10.5040/9781509932610
Jasno , S. (2015). Fu u e impe ec : Science, echnology, and he imagina ions o mode ni y. In S.
Jasano & S.-H. Kim (Eds.), D eamscapes o mode ni y: Socio echnical imagina ies and he ab ica ion
o powe (pp. 1–33). Uni e si y o Chicago P ess. h ps://doi.o g/10.7208/chicago/9780226276663.0
03.0001
Kisele a, A. (2021, July 29). Making AI’s anspa ency anspa en : No es on he EU p oposal o he
AI Ac . Eu opean Law Blog. h ps://web.a chi e.o g/web/20211006121316/h ps://eu opeanlawblo
g.eu/2021/07/29/making-ais- anspa ency- anspa en -no es-on- he-eu-p oposal- o - he-ai-ac /
23 Piasecki, Mo osoli, Helbe ge , Naud s
Lewis, S. C., Hol on, A. E., & Codding on, M. (2014). Recip ocal jou nalism: A concep o mu ual
exchange be ween jou nalis s and audiences. Jou nalism P ac ice, 8(2), 229–241. h ps://doi.o g/1
0.1080/17512786.2013.859840
Lippi, M., Con issa, G., Jabłonowska, A., Lagioia, F., Mickli z, H.-W., Palka, P., Sa o , G., & To oni, P.
(2020). The o ce awakens: A i icial in elligence o consume law. Jou nal o A i icial In elligence
Resea ch, 67, 169–190. h ps://doi.o g/10.1613/jai .1.11519
Madiega, T. (2023). Gene a i e AI and wa e ma king [B ie ing]. Eu opean Pa liamen a y Resea ch
Se ice. h ps://www.eu opa l.eu opa.eu/ hink ank/en/documen /EPRS_BRI(2023)757583
Mickli z, H.-W., Pałka, P., & Panagis, Y. (2017). The empi e s ikes back: Digi al con ol o un ai
e ms o online se ices. Jou nal o Consume Policy, 40(3), 367–388. h ps://doi.o g/10.1007/s1060
3-017-9353-0
Monze , C., Moelle , J., Helbe ge , N., & Eskens, S. (2020). Use pe spec i es on he news
pe sonalisa ion p ocess: Agency, us and u ili y as building blocks. Digi al Jou nalism, 8(9),
1142–1162. h ps://doi.o g/10.1080/21670811.2020.1773291
Naud s, L., Dewi e, P., & Ausloos, J. (2022). Meaning ul anspa ency h ough da a igh s: A
mul idimensional analysis. In E. Kos a, R. Leenes, & I. Kama a (Eds.), Resea ch handbook on EU da a
p o ec ion law (pp. 530–571). Edwa d Elga Publishing. h ps://doi.o g/10.4337/9781800371682.00
030
Newman, N., Fle che , R., Eddy, K., Robe son, C. T., & Nielsen, R. K. (2023). Digi al news epo 2023
[Repo ]. Reu e s Ins i u e o he S udy o Jou nalism. h ps:// eu e sins i u e.poli ics.ox.ac.uk/si es/
de aul / iles/2023-06/Digi al_News_Repo _2023.pd
News alues and newswo hiness. (2018). In Ox o d esea ch encyclopedia o communica ion. Ox o d
Uni e si y P ess. h ps://doi.o g/10.1093/ac e o e/9780190228613.013.850
OpenAI. (2022). Sha ing & publica ion policy [Policy o e iew]. h ps://openai.com/policies/sha ing-
publica ion-policy
Piasecki, S. (2023). Expe pe spec i es on GDPR compliance in he con ex o sma homes and
ulne able pe sons. In o ma ion & Communica ions Technology Law, 32(3), 385–417. h ps://doi.o g/1
0.1080/13600834.2023.2231326
Piasecki, S., & Chen, J. (2022). Complying wi h he GDPR when ulne able people use sma
de ices. In e na ional Da a P i acy Law, 12(2), 113–131. h ps://doi.o g/10.1093/idpl/ipac001
P i i, K., K ijge , J., Thuis, T., & S amhuis, E. (2023). F om bila e al o ecosys emic anspa ency:
Aligning GDPR’s anspa ency obliga ions wi h he Eu opean digi al ecosys em o us . In S.
Kuhlmann, F. De G ego io, M. Fe mann, H. O e dinge , & A. Se kow (Eds.), T anspa ency o opaci y:
A legal analysis o he o ganiza ion o in o ma ion in he digi al wo ld (Vol. 1, pp. 115–140). Nomos. h
ps://doi.o g/10.5771/9783748936060
Qui ing, O., Ziegele, M., Scheme , C., Jackob, N., Jakobs, I., & Schul z, T. (2021). Cons uc i e
skep icism, dys unc ional cynicism? Skep icism and cynicism di e en ly de e mine gene alized
media us . In e na ional Jou nal o Communica ion, 15, 3497–3518. h ps://ijoc.o g/index.php/ijoc/a
icle/ iew/16127
Regula ion 2024/1689. (2024). Regula ion (EU) 2024/1689 o he Eu opean Pa liamen and o he
Council o 13 June 2024 laying down ha monised ules on a i icial in elligence and amending
Regula ions (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139
and (EU) 2019/2144 and Di ec i es 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (A i icial
24 In e ne Policy Re iew 13(4) | 2024
