A prohibited practice is an AI use that must stop, not one that can proceed under ordinary controls. For example, inferring employees’ emotions from facial expressions and voice patterns during workplace video calls is banned unless the use is for medical or safety reasons; a vendor’s “engagement analytics” label does not change the use. section 7.2.2
For social scoring, check how a score is built and how it affects the person:
- A ban may apply when behavior is scored and the score is used in a different area of life from where the data came from. section 7.2.2
- A ban may also apply when the resulting disadvantage is disproportionate to the behavior. Broad character scores built from scattered signals can lead to exclusion from services. section 7.2.2
Table 7.3 Prohibited AI practices
|
Practice
|
What it
bans
|
Exception
|
|
Harmful
manipulation
|
AI using
subliminal or deceptive techniques to distort behavior and cause significant
harm
|
Threshold:
technique must cause significant harm to trigger
|
|
Exploitation
of vulnerabilities
|
AI targeting
age, disability, or socioeconomic vulnerability to distort behavior
|
None
|
|
Social
scoring
|
Scoring
people based on social behavior, where scores lead to detrimental treatment
in unrelated contexts
|
Sector-specific
scoring using data relevant to the purpose (credit history for credit
decisions, claims history for insurance)
|
|
Predictive
criminality
|
Profiling
individuals to predict criminal behavior based solely on personal
characteristics
|
AI
augmenting human assessments using objective facts linked to criminal
activity
|
|
Untargeted
facial scraping
|
Building
facial recognition databases by scraping images from the internet or CCTV
|
None.
|
|
Workplace
and education emotion recognition
|
Inferring
emotions of individuals in workplaces or educational settings
|
Medical or
safety reasons only (narrowly interpreted)
|
|
Biometric
categorization by sensitive attributes
|
Using
biometric data to infer race, political opinions, religion, trade union
membership, sex life, or sexual orientation
|
Labeling or
filtering lawfully acquired datasets
|
|
Real-time
biometric ID in public spaces
|
Live facial
recognition by law enforcement in publicly accessible spaces
|
Only for
specific crimes, missing persons, or imminent threats; requires judicial
authorization
|
Scan the listed prohibited practices and exceptions against an AI inventory. section 7.2.2
