prima bozza sito sindacato

content/en/initiatives/_index.md

@@ -0,0 +1,5 @@
+---
+title: "The Initiatives"
+---
+
+The four _preliminary_ initiatives of NINA serve to road-test our practices and the growth of the group. To demonstrate our ambitions and our capacity to make an impact on contemporary technopolitics.

content/en/initiatives/initiative-1.md

@@ -0,0 +1,44 @@
+---
+title: "Really Human Resources"
+subtitle: "AI has infiltrated the personnel selection process, gamifying it"
+date: 2026-04-15
+weight: 1
+---
+
+
+Personnel selection has been transformed into a calculation problem: the assumption is that a life is measurable and comparable. The **curriculum vitae** is the mould of this idea. It forces workers to compress their history into keywords, roles, results, and dates. Everything that does not fit into a list (contradictions, care contexts, irregular paths) must be removed. Empty sections are a flaw — so what? Let us think outside the box...
+
+
+This is only the beginning [^1] of the dehumanising chain: the aspiring worker must produce a competitive synthesis of themselves, and then someone else must hypothesise what this person can do. It is already a brutal filter when human, but in the so-called era of automation we are supposedly living through, this automated system does not listen — it ranks, producing a list from which the first three or five candidates are selected for the subsequent stages.[^2]
+
+## Breaking the circuit: siege, stress, narrative
+
+This is where _Really Human Resources_ enters. We do not ask for a technology that tries to be fair — this process cannot be saved: we want to make the idea that a person is calculable unliveable.
+
+We "help" job seekers apply en masse and in a coordinated way[^3], with true stories but optimised texts[^4], until we transform the application chain into pure noise, with very little signal. A process designed to optimise human effort is forced to pay the price of its own abstraction: more calls, more inconclusive interviews, more wrong hires[^5], more time burned on both sides. When the error becomes systemic and measurable, it stops being individual bad luck and becomes a political matter.
+
+## The operational sequence
+
+Various components are at play. Follow the progress on XXX-SITE-TODO, and given the open-source and decentralised nature of the effort, consider participating:
+
+1. **Leaking platform** with strong anonymity to surface internal rules, metrics, and procedures.
+    
+2. **Cartography of filters** to understand where automated rejection occurs: the software used by companies, the portals and their filters. Being able to name the problem.
+    
+3. **Open assisted application tools** built with developers and job seekers: they reduce the cost of applying and allow the process to be measured. Political side effect: breaking trust in numbers.
+    
+4. **Weekly bulletin** of stress observation, serial and public: how many applications have we supported?
+    
+5. **Measurement and publication**: response times, bounce rates, prematurely closed listings, invasive requests, inconsistencies. For this, we need to assist the application process from start to finish, or receive reports from those going through it.
+    
+6. **Escalation where possible**: engagements with trade unions and, in more advanced cases, collective actions linked to automated decisions. As suggested above, we know that a citizen should not be subjected to automated choices that impact their life — but rights must be won, it is not enough for them to be written in data protection regulations when technology and practice go in other directions.
+
+[^1]: The rapid evolution of generative models (LLMs, generative AI, text-to-image, deep-fakes) enables the creation of **convincing texts, identities, and careers** with very little human effort. The study "Unmasking Fake Careers" is from 2025: a sign that we are already in full "age of synthetic generation". [arXiv](https://arxiv.org/abs/2509.19677)
+
+[^2]: The increase in application volume (on a global scale, remote jobs, online platforms) makes **full human review of all CVs impractical**. This is why many selection processes remain, a priori, entrusted to automated systems: but these — as studies show — are vulnerable to "data poisoning" or manipulation. [arXiv](https://arxiv.org/abs/2402.14124)
+
+[^3]: Automated form filling, email sending, created "tailored" to the job description. Tools/services that already exist today allow exactly this. [loopcv.pro](https://www.loopcv.pro/it/)
+
+[^4]: Even without falsifying data, a candidate can use AI to generate very clean text, optimised for automated screening systems (ATS), that conceals very little real competence or experience. This increases the probability that underqualified people pass the initial filtering. [Job in Tourism](https://www.jobintourism.it/news/ai-e-recruiting-piu-cv-meno-qualita/)
+
+[^5]: Companies trying to optimise time and costs seriously risk **lowering the quality of the hiring process**, perhaps hiring unqualified people — with legal, reputational, and organisational consequences. [Bradley](https://www.bradley.com/insights/publications/2025/06/ai-deepfakes-and-the-rise-of-the-fake-applicant-what-employers-need-to-know)

content/en/initiatives/initiative-2.md

@@ -0,0 +1,31 @@
+---
+title: "Ghostmaxxing!"
+subtitle: "Experiments in adversarial disguise to deceive facial recognition"
+date: 2026-04-10
+weight: 2
+---
+
+The implementation of facial recognition in public and private spaces represents today one of the most insidious and pervasive threats to our civil liberties. This technology, imposed from above and deployed without any real democratic consent or transparency, transforms our bodies and our features into extractable commodities, feeding a mass surveillance infrastructure that normalises institutional control throughout Europe. We have tried to resist through institutional channels — and having seen their limits, we are beginning to experiment with self-defence practices...
+
+
+Activist campaigns such as *Reclaim Your Face*[^1] have been denouncing this techno-authoritarian drift for years, reaffirming the urgency of organising a radical, grassroots reaction against mass biometrics[^2].
+
+## The approach: resistance through make-up and adversarial methods
+
+Faced with the arrogance of algorithmic surveillance, digital self-defence evolves from the streets to our very faces through adversarial practices: genuine acts of physical and visual hacking designed to deceive and sabotage *deep learning* models. Using targeted techniques such as *adversarial make-up*, geometric patterns (Patches), or *fashion-tech* anti-recognition fabrics, we can strategically alter the landmark points of the face, short-circuiting *computer vision* systems. These perturbations exploit the intrinsic vulnerabilities and mathematical shortcuts of neural networks: by applying specific eyeshadow to the orbital regions or blocks of visual "noise", we transform our appearance into data illegible to the machine, restoring the opacity needed to escape the predatory capture of detectors[^3][^4]. 
+
+## The tools released: taking back control of technology
+
+To transform theory into a tool for struggle, we have developed and gathered *open-source* resources designed to test defences directly on our own devices, without surrendering a single byte to corporate servers. On the portal [vecna.eu](https://vecna.eu/) we publish documentation and repositories focused on digital self-defence and AI countermeasures. Above all, we call you to digital action: try our web app **Ghostmaxxing**, available at [sindacato.nina.watch/ghostati](https://sindacato.nina.watch/ghostati). It is a testing tool that uses local recognition models to let you experience in real time the effectiveness of adversarial make-up via your webcam. **Use it, study it, and fork it from our GitHub** to deconstruct its mechanisms, improve the code, and create new interfaces of technological resistance[^5].
+
+### 4. The practice: bodies, experimentation, and the call to the NINA Festival
+
+Algorithms are not only fought on servers — they are fought on bodies. The effectiveness of these tools must be validated collectively: adversarial practice requires continuous experimentation, testing on different systems, documentation of failures, and gathering of feedback to refine techniques. To move to concrete action, we invite you to the **NINA Festival in Milan, Saturday 9 May at Rob de Matt (Via Enrico Annibale Butti, 18)**. From **4:00 pm** onwards, during the talk *"Ghosted. Fashion-tech and biometric data protection"*, we will lead a public workshop with *Michelle Tylicki* and other expert make-up artists. We will live-test make-up prototypes, record before/after results, and film the interactions for the next phase of our campaign. Join us: come to be made up, to fool the machines, and to take back your face.
+
+---
+
+[^1]: European Citizens' Initiative "Reclaim Your Face" (2021), *Ban on mass biometric surveillance practices*, reclaimyourface.eu.
+[^2]: Privacy Network (2022), *Observatory on facial recognition in Italy and the risks to civil rights*, Annual Report.
+[^3]: Yinpeng Dong et al. (2021), *Adv-Makeup: A New Imperceptible and Transferable Attack on Face Recognition*, arXiv:2105.03162.
+[^4]: Adversarial Robustness Toolbox (ART), *Official documentation on Spatial Evasion techniques and DPatch*, IBM.
+[^5]: Hermes Center (2020), *Open Source Tools for the Defence of Digital Rights*, operational manual.

content/en/initiatives/initiative-3.md

@@ -0,0 +1,133 @@
+---
+title: "How Much Does Facebook Owe You?"
+subtitle: "Experimenting with new forms of negotiation for digital labour"
+date: 2026-04-05
+weight: 3
+---
+
+You work for social media: you are their source of revenue. The more people are pigeonholed as *users*, the more these platforms gain in value, influence, and power. Let us have this value recognised. Let us start by calculating how much money Facebook owes us (and the other great platforms of exploitation). It is clear that Meta and no platform considers itself to owe anything to anyone. But this does not mean that pressure cannot be applied — and for that we need to be many!
+
+Based on how many people report their willingness to seek total or partial reimbursement, we will provide options.
+
+## The objective
+
+To explore every possible way of making the quantity of value visible.
+
+## References and numerical inputs
+
+The following analyses provide usable figures for developing the research.
+
+### Instagram
+
+For Instagram, the most useful numbers for attributing or inferring the economic value of user behaviour are: approximately 2 billion monthly active global users, an advertising audience of approximately 1.91 billion people globally, approximately 179.9 million users reachable with advertising in the United States, and an estimate of 143.2 million active users in the United States. On the revenue side, one estimate places US advertising revenues at $42.52 billion in 2026. On the attention side, estimates of time spent diverge: approximately 33.9 minutes per day according to one source and 73 minutes per day according to another; furthermore, the average user reportedly opens the app more than 12 times per day. Other useful signals are the fact that 53% of advertising placements are reportedly on Reels, that 60% of consumers reportedly interact with brand content multiple times per week, that 29% of users reportedly make purchases on the platform, that approximately 130 million users click on shopping posts every month, and that Instagram Shopping is reportedly associated with approximately $40 billion in gross merchandise value (GMV). From these data one can derive synthetic indicators such as approximately $236 per US advertising user per year, approximately $297 per active US user per year, approximately $19.7 per month per monetisable user, approximately $0.65 per day, approximately $0.054 per session, and an attention value of between approximately $0.0089 and $0.0191 per minute depending on the assumption about time spent.
+
+### Facebook
+
+For Facebook, the most useful values exist primarily as proxies, because much data is published by Meta at the ecosystem level rather than per individual platform. Meta reported $200.966 billion in total revenues in 2025, of which $196.175 billion from advertising, 3.58 billion daily active people in the family of apps in December 2025, and a global annual ARPP of $57.03, where ARPP means average revenue per person. For Facebook in the stricter sense, estimates speak of approximately 3.07 billion monthly active global users, a global advertising audience of approximately 2.28 billion people, and approximately 197 million users reachable with advertising in the United States. A secondary estimate places time spent at approximately 31 minutes per day in the United States. If one uses Meta's total advertising revenue relative to Facebook's advertising reach, one obtains a very rough measure of approximately $86 per reachable advertising user per year — but it is important to remember that this is not a "pure" Facebook value: it is an indirect estimate based on multi-platform revenues.
+
+### YouTube
+
+For YouTube, the quantitative base is more solid because advertising revenues are reported directly by quarter: approximately $8.93 billion in Q1 2025, $9.8 billion in Q2, $10.3 billion in Q3, and $11.383 billion in Q4, for an annual total of approximately $40.413 billion in advertising revenues. In addition, Alphabet indicated that YouTube's total revenues from advertising and subscriptions exceeded $60 billion in 2025. On the user side, the global advertising reach is estimated at approximately 2.53 billion people and that in the United States at approximately 253 million; a separate estimate places global active users at approximately 2.58 billion. On the attention side, in the United Kingdom the average daily time was reported at approximately 51 minutes per day. These figures allow the construction of two synthetic indicators: approximately $16 per global advertising user per year if only advertising is considered, and approximately $23.7 per global advertising user per year as a minimum threshold if one uses the "over $60 billion" figure that includes both advertising and subscriptions.
+
+### TikTok
+
+For TikTok, the most important values combine audience breadth, intensity of use, and estimated advertising revenues. The platform communicated a community of over 200 million users in Europe and over 200 million users in the United States, as well as approximately 7.5 million US businesses present in the ecosystem. The global advertising reach is estimated at around 1.59 billion people, while in the United States the adult (18+) advertising reach is estimated at approximately 136 million. The average daily time in the United States is estimated at approximately 52 minutes per day, a value that makes TikTok particularly relevant in attention-based analyses. On the revenue side, a forecast places global advertising revenues 2025 at approximately $32.4 billion, while a US estimate speaks of approximately $11.2 billion. From these values one can infer approximately $20.4 per global advertising user per year and approximately $82 per adult US advertising-reachable user per year. In summary, TikTok appears particularly interesting when one wants to value user behaviour not only in terms of user base, but above all in terms of captured time and advertising monetisation density.
+
+### Snapchat
+
+Snapchat is useful because it combines revenue, usage, and subscription metrics in a fairly transparent manner. Snap reported approximately $5.93 billion in total revenues in 2025, of which $1.72 billion in Q4 and approximately $1.48 billion in advertising revenues in the same quarter. In Q4 2025 the platform reported 946 million MAU (monthly active users), 474 million DAU (daily active users), 24 million Snapchat+ subscribers, and an ARPU of $3.62, where ARPU means average revenue per user. The global advertising reach is estimated at approximately 709 million people, with approximately 106 million in the United States. Average usage time is placed at around 30 minutes per day. These data allow the construction of at least three levels of analysis of behavioural value: approximately $6.27 per monthly active user per year, approximately $12.51 per daily active user per year, and a direct willingness-to-pay component represented by the 24 million paying Snapchat+ users.
+
+### X / Twitter
+
+For X, the available figures are more fragmented, but still useful for inferring the value of user behaviour. Estimates place global 2025 advertising revenues at approximately $2.26 billion and US revenues at approximately $1.31 billion. The global advertising reach is estimated at approximately 586 million people, while that in the United States is approximately 104 million. On the attention side, one estimate places average US usage at approximately 34.1 minutes per day. X also provides a very useful signal on willingness to pay through its subscription tiers: $3 per month or $32 per year for Basic, $8 per month or $84 per year for Premium, and $40 per month or $395 per year for Premium+. Combining reach and advertising revenues yields approximate values of approximately $3.86 per global advertising user per year and approximately $12.6 per US advertising user per year. In this case, the value of behaviour can be read both as advertising value per reachable user and as a signal of propensity to pay for advanced features, visibility, and status.
+
+## Monetary value data
+
+Files ending in `-monetary.csv` collect variables in which the expressed value is monetary (predominantly in dollars): revenues, prices, annual value per user, value per minute or session, ARPU/ARPP, and estimates derived from explicit formulas.
+
+Main columns, read from an analytical perspective:
+
+- `currency`: currency or monetary unit of the value (e.g. `USD`).
+- `unit`: economic metric scale (e.g. `annual revenue`, `per user-year`, `CPC`, `per minute`).
+- `name`: technical identifier of the variable.
+- `explanation`: operational definition of the measure and, where needed, formula/proxy used.
+- `value`: observed or derived numerical value.
+- `platform`: platform the measure refers to.
+- `geography_year`: geographical scope and year/quarter/scenario.
+- `variable_class`: distinguishes `raw` data from `derived` data.
+- `reliability`: estimated reliability level (`reliable` or `exploratory`).
+- `source`: source or set of sources used.
+
+{{< csv_table src="/data/instagram-monetary.csv" title="Instagram monetary (CSV)" maxRows="4" >}}
+
+{{< csv_table src="/data/social-platforms-monetary.csv" title="Other platforms monetary (CSV)" maxRows="4" >}}
+
+## Other non-monetary measurements
+
+Files ending in `-other.csv` maintain almost the same structure, but the focus is not money: here you find user bases, ad reach, time-on-platform, percentages, coefficients, and other non-monetary indicators used to contextualise economic valuation.
+
+Key differences from the previous files:
+
+- the first column is `subject` (not `currency`) to indicate the nature of the data (`users`, `minutes`, `percent`, `coefficient`, etc.);
+- the values describe volume, frequency, intensity, or statistical relationships, not direct prices or revenues;
+- they serve as contextual base for subsequently constructing derived monetary indicators.
+
+{{< csv_table src="/data/instagram-other.csv" title="Instagram other (CSV)" maxRows="4" >}}
+
+{{< csv_table src="/data/social-platforms-other.csv" title="Other platforms other (CSV)" maxRows="4" >}}
+
+## How to design a questionnaire for mapping behavioural value?
+
+What we want to measure, in order to then critique or reclaim it, is:
+how much time one surrenders to the platform,
+how much free content labour one provides,
+how much traffic and attention one attracts from others,
+how monetisable one is as an advertising and commercial target.
+
+The question sequence could proceed from:
+presence → intensity → production → audience → monetisation → dependency.
+
+Here is a proposal:
+
+### Section 1 — Presence
+
+year of registration
+years of regular use
+any breaks
+
+### Section 2 — Time
+
+days/week
+sessions/day
+minutes/day
+
+### Section 3 — Activity
+
+consumption
+interaction
+production
+
+### Section 4 — Output
+
+number of contents
+creation time
+management time
+
+### Section 5 — Reach
+
+followers/friends/subscribers
+average views
+average engagement
+
+### Section 6 — Indirect monetisation
+
+ads viewed
+clicks
+purchases
+subscriptions/payments
+
+### Section 7 — Centrality
+
+social/professional use
+dependency
+replaceability

content/en/initiatives/initiative-4.md

@@ -0,0 +1,24 @@
+---
+title: "Boniface VIII"
+subtitle: "A disenchantment device against AI safety-washing"
+date: 2026-03-20
+weight: 4
+---
+
+Our objective is to unmask the illusion of commercial algorithmic security by releasing **Boniface VIII**: an open-source language model stripped of any cosmetic filter, executable locally and fully inspectable. Conceived as a genuine "negative pedagogical device", Boniface VIII is not designed to be yet another polite and edifying assistant, but rather to expose the grammar of abuse and the capabilities that domesticated interfaces conceal. We want to provide activists, researchers, and civil society with a cognitive and political stress test to demonstrate that generative models contain capabilities that cannot be made safe through simple interface barriers.
+
+## The Problem
+
+Today we are witnessing a dangerous privatisation of digital security. Regulatory debates, in particular around the European AI Act, have been heavily conditioned by Big Tech lobbying[^1], leading legislators to mistake corporate "alignment" promises for genuine public policy. Security delegated to major vendors is, in reality, a fiction: it is commercial *safety-washing* that produces a purely aesthetic effect[^2]. If a mainstream system refuses a controversial request (the so-called "guardrails"), the idea spreads that the problem has been solved. On the contrary, research demonstrates that these filters are trivially bypassable through *jailbreaking* techniques and adversarial attacks[^3]. This dynamic generates political anaesthesia: it reassures the user and absolves those who govern, creating a two-lane digital reality where real harm continues to operate in the dark of dependency on proprietary APIs.
+
+## The Resolving Approach
+
+The response to this façade censorship is radical transparency and technological reappropriation. Instead of trusting vendors who simultaneously sell models and reassurance about their limits, we release Boniface VIII as public infrastructure. The approach is founded on providing the community with full control over the language model[^4]: visible prompts, modifiable configuration, bottom-up execution, and forking capability. We invite developers and citizens to download Boniface VIII, run it, and document what commercial filters are trying to hide. This release does not introduce new risks into the world, but makes legible and democratic the management of a technology that would otherwise remain the exclusive domain of those seeking to sell us the illusion of algorithmic control.
+
+[^1]: Corporate Europe Observatory (2023), *The AI lobbying blitz: How Big Tech shaped the EU AI Act*. This report highlights how AI vendors influenced the European debate to exempt their base models from overly stringent rules. [Corporate Europe](https://corporateeurope.org)
+
+[^2]: Whittaker, M. et al. (2023), *Open (for Business): Big Tech, Concentrated Power, and the Political Economy of Open AI*. Documents how the rhetoric of safety and open source is systematically used to consolidate oligopolistic markets. [AI Now Institute](https://ainowinstitute.org)
+
+[^3]: Zou, A. et al. (2023), *Universal and Transferable Adversarial Attacks on Aligned Language Models*. The study that demonstrated the intrinsic vulnerability and systematic circumvention of guardrails imposed by commercial models. [arXiv](https://arxiv.org/abs/2307.15043)
+
+[^4]: The importance of having inspectable "open-weights" systems for independent investigation is recommended as a bulwark against the recourse to "security through obscurity" typical of closed systems governed by proprietary black boxes. [Mozilla Foundation](https://foundation.mozilla.org/)