The European approach to AI is frequently marketed as a third way, an alternative to both an AI-powered surveillance capitalism and an AI-powered surveillance state. Landmark instruments such as the AI Act, the General Data Protection Regulation and the Digital Services Act are held up as proof that Europe can foster trustworthy AI while safeguarding democratic values. The word while does heavy lifting here, signaling a compromise between value systems that are supposedly reconciled rather than traded off. Knopp argues that this compromise is not merely a policy stance but a mode of justification, one that selectively absorbs some criticisms while excluding others in order to establish the moral conditions under which data extraction becomes possible at all.

The case at the heart of the study is the Hamburg police’s trial of automated behavioral recognition on Hansaplatz, a square in a predominantly migrant neighborhood undergoing gentrification. The system relies on algorithmic pose estimation: digital video frames are converted into vector graphs, producing what proponents call digital skeletonization, a normalized, decontextualized and depersonalized representation of human bodies. Unlike facial recognition, the software strips away body size, skin color, hair, clothing and body shape, then classifies movements against a predefined set of police-relevant activities such as beating, kicking or defensive posture. A three-month proof-of-concept ran in 2023, and by February 2025 the city parliament had passed the first German law explicitly permitting police to transfer both anonymized and non-anonymized surveillance data to external partners for machine learning.

Getting to that point required enormous sociotechnical investment, what the study, drawing on the pragmatist sociology of Laurent Thévenot, calls form investments. The square itself had to be engineered into something like a laboratory, with AI-ready cameras, an interoperable CCTV system, high-security data infrastructure and an unobstructed sightline across the surveilled space. CCTV operators had to learn to label the system’s outputs, maintain performance logs and interpret the different ways it produces false alarms. Because public spaces rarely yield sufficient footage of atypical behavior, police in Mannheim, where the system was first implemented, even staged show fights in front of cameras, with riot police enacting scripted violence to generate training data. These theatrical performances, the study notes, risk inscribing selective police assumptions about what violence looks like directly into datasets and classification algorithms.

Knopp’s central analytical move is to name these costs data sacrifices. Datafication, he argues, is never a neutral capture of reality but a costly operation of appropriation, exclusion and reductive generalization, one that abstracts traces of events from the complex situations in which they occur. The concept does double duty: it highlights the material and social expenses of formatting the world into data, and it foregrounds the critical agency of social actors who point out what has been disregarded. Where the influential notion of the boundary object emphasized dynamic compromises between local knowledges, the data sacrifice directs attention to the moral work that presents these efforts and exclusions as worthwhile contributions to a higher common good.

Using Critical Discourse Analysis of 57 parliamentary documents spanning from the conservative opposition’s first request in May 2019 to the passage of the new law in February 2025, the study reconstructs five suborders of worth through which the system was justified and attacked: security, individual freedom, social justice, automation and experimentalism. The conservative CDU opened the debate with a catastrophic picture of declining safety, describing citizens who hardly dare to walk alone in the dark and squares that had become crime hotspots plagued by fights and knife attacks. Such narratives perform a dual function, establishing the legal grounds for CCTV while constructing moral urgency around the high discursive value of security.

What made the AI project politically viable, however, was not the old zero-sum logic in which freedom is traded for security. Instead, proponents framed automation itself as the compromise. Conventional CCTV was criticized as inefficient, with one Social Democrat noting that officers have to watch twenty screens all day long, while AI promised to automate attentiveness, cut costs and enable real-time preemption. More remarkably, the technology was sold as a civil-liberties upgrade: because pose estimation supposedly erases personal traits, the innocent bystander remains out of sight, is not stored in any records, and, in the words of one conservative parliamentarian, has nothing to fear. This framing allowed the Green Party, historically rooted in surveillance-skeptic social movements, to reposition itself as a supportive but watchful partner, declaring itself open to the outcomes of the trial while endorsing its potential.

The same logic was extended to social justice. Because the stick figures produced by skeletonization carry no age, gender or ethnicity, proponents claimed the system could not discriminate, positioning it as superior both to human operators and to facial recognition, which had once made the identical promise. Critics from the Left Party saw a discriminatory trap instead, arguing that people with physical disabilities whose movements differ from those classified as normal, or homeless people who sleep rough, would be flagged as atypical, and that discrimination begins the moment someone is captured and unjustly observed, not when police intervene. Rather than finding common ground, these critics were excluded from legitimate debate, dismissed as ideological, ignorant or dishonest by parties that otherwise shared the same civic vocabulary.

When present-day failures could not be justified by the system’s actual capabilities, proponents turned to experimentalism, the solutionist order of worth that frames innovation as an end in itself and an external compulsion. Hamburg was portrayed as lagging behind other federal states; AI was declared to be everywhere, making it quite unrealistic not to exploit it; and German-developed software was celebrated as a matter of competing with American and Chinese surveillance markets. Trial and error became the sole pathway to progress, justifying security risks, costly infrastructure and intrusions into civil rights as temporary sacrifices on the road to an imagined future in which all contradictions would finally be resolved. The study argues that this radical temporalizing of moral paradoxes turns the object of criticism into a moving target and undermines public accountability, since evaluations of present failure are always deflected toward a not-yet-existing entity designed to withstand critique.

The paradoxes culminated in the new Hamburg law, which authorizes the transfer of citizen surveillance data to external developers for machine learning and has since been adopted in other German states. Knopp’s conclusion is pointed: the third way does not simply balance innovation and rights, it equips social control and data extraction with moral underpinnings, internalizing select civic values and critics while excluding others, until the industrialization of surveillance itself becomes a civic value. As local AI trials proliferate across Europe, the study suggests that the decisive battles over what algorithms may see, what data they may consume and who pays the price will be fought not in Brussels policy papers but in precisely such local disputes, where the sacrifices are made concrete and the justifications, for those willing to look, become visible.

Subject of Research: Parliamentary justification and critique of data extraction in an AI-based CCTV surveillance trial in Hamburg, Germany

Article Title: Data sacrifices and the ‘third way’ toward AI: justification and critique in local conflicts over automated surveillance

Article References: Knopp, P. (2026). Data sacrifices and the ‘third way’ toward AI: justification and critique in local conflicts over automated surveillance. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03075-w

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03075-w

Keywords: artificial intelligence, surveillance, datafication, CCTV, police, Hamburg, third way, critical data studies, pose estimation, machine learning, parliamentary discourse, civil liberties

The study tackles a persistent methodological problem in the surveillance literature. Most previous evaluations have measured camera effects within discrete spatial units, such as fixed-radius buffers around each camera or administrative boundaries like census tracts. Those approaches implicitly assume that the deterrent influence of a camera stops abruptly at an artificial border, which is implausible in practice. To avoid that artificiality, Kim and Wo adopted the egohood framework developed by John Hipp and Adam Boessen, which constructs overlapping, non-discrete neighborhoods centered on every census block. Each quarter-mile egohood contains portions of the buffers of neighboring blocks, so spatial influence can flow continuously across space rather than being clipped at rigid edges.

The quarter-mile radius was not arbitrary. The researchers estimated models across egohood radii ranging from 0.10 to 1.00 miles in small increments and found a clear spatial threshold: the negative association between CCTV density and crime reached its greatest magnitude, stability and statistical significance at roughly 0.25 miles, then steadily weakened and became imprecise beyond half a mile. That distance also matches established findings on American pedestrian behavior, where most adults are willing to walk about a quarter to half a mile rather than drive. The implication is that the deterrent power of cameras is intensely local, tied to the immediate environment a person moves through rather than the broader district.

To measure outcomes, the authors combined four datasets: crime incidents reported to the Metropolitan Police Department in 2018, geocoded at the 100-block level; the police department’s neighborhood camera locations updated in 2018; 2010 Census data describing socioeconomic and demographic conditions; and 2015 land parcel data from the District Department of Transportation. Crimes were aggregated into two categories, with violent crime comprising homicide, robbery and aggravated assault, and property crime comprising burglary, motor vehicle theft and larceny. CCTV presence was captured both as a binary indicator of whether at least one camera existed in each egohood and as a continuous count of cameras, allowing the team to distinguish the effect of having any surveillance from the effect of camera density.

The baseline results, estimated with negative binomial regression models that account for the over-dispersed count nature of crime data, were consistent and modest in size. The presence of neighborhood CCTVs was associated with roughly a 4.5 percent reduction in violent crime and a 9.4 percent reduction in property crime. Each additional camera within an egohood corresponded to about a 3.7 percent lower violent crime count and an 8.2 percent lower property crime count. These estimates held after controlling for poverty, racial composition and heterogeneity, homeownership, length of residence, housing occupancy, age structure, immigrant concentration and the proportions of land devoted to hotels, stores, malls, supermarkets and restaurants. The maps underlying the analysis showed crime concentrated downtown, around Dupont Circle and Logan Circle, and in parts of Georgetown and other mixed-use nightlife areas.

The most striking contributions come from the moderation analyses, in which the researchers interacted camera counts with each structural and land use variable individually and applied the Benjamini-Hochberg procedure to guard against false discoveries across 26 interaction tests. For violent crime, cameras were more effective in low-poverty neighborhoods, where additional cameras produced the steepest declines in offending. For property crime the contrast was sharper still: in high-poverty egohoods the camera-crime line was essentially flat, indicating no measurable benefit, while in low-poverty areas the association was clearly negative. The authors suggest that structural disadvantage may undermine the deterrent reach of formal surveillance, and that cameras cannot substitute for social investment in communities facing poverty, inequality and weakened cohesion.

Homeownership produced a complementary pattern. CCTV showed its strongest effects in neighborhoods with low levels of homeownership, where many residents rent and informal social control is likely weaker. In areas with medium to high homeownership, typically more residentially stable places, additional cameras yielded minimal to negligible benefit, as indicated by flat slopes in the interaction plots. The interpretation follows from social disorganization theory: in stable neighborhoods, homeowners may already sustain robust guardianship, collective efficacy and social capital, so formal surveillance adds little. In renter-dominated areas, cameras may compensate for thinner informal control networks, functioning as virtual guardians where natural surveillance by residents is less established.

Land use also shaped effectiveness. The crime-reducing effect of cameras was most pronounced in egohoods with high proportions of supermarket and restaurant land use. These commercial areas generate heavy foot traffic and elevated opportunities for crime, since motivated offenders and suitable targets converge there, but they also benefit disproportionately from added surveillance. In neighborhoods with low to medium densities of supermarkets or restaurants, increasing camera counts produced little measurable change in either violent or property crime. From a criminal opportunity perspective, the authors argue, cameras act as capable virtual guardians precisely in those high-activity settings where perceived detection risk matters most to a potential offender weighing the costs and rewards of offending.

The findings arrive amid broader evidence that CCTV works, but modestly and conditionally. The study aligns with a large meta-analysis synthesizing 76 evaluations over four decades, which found a statistically significant but small overall crime reduction, stronger effects when cameras were actively monitored and paired with complementary strategies, and the largest benefits in parking facilities. Earlier work in Cincinnati similarly found significant reductions in assault, robbery and burglary only in residential settings. The D.C. study extends this conditional view by showing that effectiveness is not only setting-specific but also structured by the socioeconomic fabric and physical form of the places where cameras are installed, from renter-heavy blocks to busy commercial strips.

The authors caution that their design has limits. They could not directly test the mediating mechanisms, such as informal social control or collective efficacy, that theoretically explain the patterns they observed; crime outcomes and covariates are separated by a temporal lag inherent in available block-level Census data; and the analysis captures a single city at a single point rather than a longitudinal or quasi-experimental design that would strengthen causal inference. Future work, they suggest, should examine time-of-day and seasonal variations in camera effects, and should replicate the egohood approach in other cities. Even so, the practical message is clear: deploying cameras uniformly across a city will produce uneven returns, and surveillance policy should be tailored to the social and physical character of each neighborhood rather than treating every block as the same.

Subject of Research: The neighborhood-contingent effectiveness of public CCTV surveillance in reducing violent and property crime

Article Title: A Spatial Examination of the Presence of CCTV Cameras and Neighborhood Crime

Article References: A Spatial Examination of the Presence of CCTV Cameras and Neighborhood Crime. (n.d.). https://doi.org/10.1007/s12103-026-09936-w

Image Credits: AI Generated

DOI: 10.1007/s12103-026-09936-w

Keywords: CCTV, surveillance, crime prevention, neighborhood crime, Washington DC, egohoods, spatial criminology, land use, social disorganization, violent crime, property crime, urban policy