A companion deck to the article
06 / 2026
The Digital
Sovereignty Gap
Act 01
01
The data trade.
AI as the most prominent new technology of our moment, built by borrowing — and in many documented cases, taking — from the intellectual property of the people who built the existing digital record.
The premise
Information and data are the fuel of this economy. Without user input, most platforms are skeletons.
The question is not whether AI is built on existing data. It is whether the people whose data, work, voice, and likeness make AI possible have any agency over how it is being used — and whether the legal infrastructure to push back is symmetrically available to them.
Headline
$1.5B
Bartz v. Anthropic, finalised May 2026.
The largest publicly reported copyright recovery in US history, paid for piracy in the construction of training data.
Active AI copyright suits
70+ in US & international courts as of June 2026
Implied per-work rate
$3,113 per book · ~482,000 works covered
Cumulative claimed damages
$50B+ across all active AI copyright cases
First fair-use ruling
Thomson Reuters v. ROSS — Feb 2025, rejected the defense
Sources: Copyright Alliance · Axis Intelligence Research · US Copyright Office Part 3 Report
Case law · Music
The industry that moved first on AI training data.
UMG v. Udio
Settled Oct 2025
Settled with opt-in licensing. Co-launching a "walled garden" AI music platform in 2026 where artists retain control over voice, likeness, and composition.
WMG v. Suno
Settled Nov 2025
Suno launching entirely new licensed models in 2026; pre-settlement models being phased out. Artists hold "full control" over use of name, image, likeness, voice.
Sony v. Suno · Sony v. Udio
Ruling expected summer 2026
Sony has settled with neither. The fair-use ruling expected this summer will set precedent for every subsequent AI music case.
UMG / Concord / ABKCO v. Anthropic
Amended Jan 2026
$3.1B amended complaint covering 20,000+ songs. Theory: Anthropic torrented lyrics — leaning on the Bartz precedent that piracy isn't protected even if training is.
Sources: Chartlex Music Industry AI Lawsuits Tracker · Norton Rose Fulbright · Axis Intelligence
Case law · Image, video, news
The studios and publishers now in court.
NYT v. OpenAI / Microsoft
Filed Dec 2023 · ongoing
Could set the broadest precedent. Evidence of outputs that closely mirror original reporting raises the question: when does "learning from" become "copying"?
Disney / Universal v. Midjourney
Filed 2025 · ongoing
First major test of output-level copyright liability for AI image generators. Disney's behemoth status produces unique dynamics not seen in earlier cases.
Disney / Universal / WB v. Midjourney (video)
Filed 2026 · ongoing
First case involving AI video generation, not just text or images. Targets generation of recognisable copyrighted characters.
Getty v. Stability AI
UK lost on jurisdiction · US 25-cv-06891 active
12M+ photographs and metadata allegedly used in building Stable Diffusion. UK case largely lost on jurisdictional grounds; US case in earliest stages.
Sources: McKool Smith AI Infringement Tracker · Manuscript Report · CourtListener
Regulation · The state response
Where the law is catching up.
EU AI Act · Art. 50
Applicable 2 Aug 2026
Machine-readable marking of AI-generated audio, image, video, and text. Deepfake disclosure obligations on deployers. Output-side transparency.
EU AI Act · Art. 53
Full enforcement Aug 2026
Training-data summary disclosure mandate. €15M or 3% of global annual turnover for GPAI providers that fail to publish.
TAKE IT DOWN Act (US)
Signed 2025 · effective May 2026
Notice-and-removal for AI-generated non-consensual intimate imagery. 48-hour platform removal. Scope: sexual content only.
NO FAKES Act · DEFIANCE Act
Reintroduced 2025
Federal bills on unauthorised AI replicas (NO FAKES) and civil remedy for NCII deepfakes (DEFIANCE). Both pending.
Denmark · Copyright Amendment
Passed late 2025
First law treating every person's face, voice, and body as IP by default. 50-year post-death protection. Model being pushed across the EU.
US state laws
47 states · mid-2026
Patchwork: 30 with political deepfake disclosure, 47 with NCII protections, fragmented right-of-publicity coverage. Constitutional challenges ongoing.
Neural data laws
Colorado · California · Montana · UNESCO 2025
It isn't just AI. The neurotech industry is maturing alongside this. Three US states have passed neural data laws and UNESCO issued a 2025 recommendation. Global picture remains uneven.
Sources: EU Regulation 2024/1689 · SAG-AFTRA · Reuters · State of New York
Who can fight back
Those with the
infrastructure.
- Major rights holders — UMG, WMG, Disney, NYT
- Class-action plaintiffs — Bartz authors, Andersen et al.
- National regulators — EU, certain US states
- Trade unions — SAG-AFTRA, DGA, WGA
And who cannot — yet
Those without it.
- Independent artists in the Global South
- Non-union performers across every industry
- Jurisdictions with no AI case law
- Indigenous-language workers — absent from training, absent from terms
Act 02
02
Workers & the literacy gap.
The same pattern, viewed through a workforce that organised early — and the gap that opens between those inside the floor and those outside it.
SAG-AFTRA · 2023 → 2026
A workforce that understood early.
Nov 2023
2023 Strike resolution. First-ever AI provisions in the SAG-AFTRA / AMPTP TV/Theatrical Agreement.
May 2025
Commercials Contracts ratified. Strongest contractual AI guardrails achieved to date.
Jul 2025
Interactive Media Agreement. Consent and disclosure for AI digital replicas; suspension of consent during strikes. Ends nearly year-long video game strike.
Dec 2025
NY S.8391 / A.8882 signed. Postmortem replica protections championed by SAG-AFTRA.
May 2026
TAKE IT DOWN Act in effect. 48-hour removal of AI-generated non-consensual imagery.
Jun 2026
2026 TV/Theatrical Agreement ratified by 91.42%. 12 AI provisions. Expands replica protections, restricts use of synthetics.
Source: SAG-AFTRA AI Bargaining And Policy Work Timeline · Variety · IndieWire
2026 TV/Theatrical Agreement
12 AI provisions in the latest contract — what collective bargaining gets you.
01Consent required to create a digital replica
02Consent required to use a digital replica
03"Reasonably specific description" of use before deploying a synthetic
04Arbitration provision limiting AI replica use to edge cases
05Strike-suspension clause on digital replica generation
06Notice when actor data is licensed to a third party for AI training
07Security requirements for stored replica scans & data
08Specific protections for background actors
09Voice replica rules covering all in-game material
10Visual replica rules for scripted cinematic scenes
11Penalties for using synthetics in place of real performers
12Recognisability standard ("objectively identifiable")
Source: SAG-AFTRA published summary · Davis+Gilbert LLP · IndieWire
The horizon · five years
In five years, AI models will generate content that today requires an actor and a full crew.
The question is not whether that wave arrives. It is whether the people whose livelihoods it touches have the digital literacy, the institutional infrastructure, and the legal vocabulary to be in the conversation about how it is structured. Or whether it arrives, as so many things have, fully formed — with the terms already set.
Act 03
03
The framework as rubric.
A way of naming what is happening — across four pillars of participation and three levels at which sovereignty either holds or breaks down.
The pillars
Four pillars of digital participation.
P 01
Digital
Presence
Whether you are in the record. What surfaces, what doesn't. Authored, or decided for you.
P 02
Digital
Literacy
Knowing what the systems do, what they need from you, and what each data point you produce says.
P 03
Digital
Economy
Participation in the economic life the digital substrate now carries. Earnings, IP, contracts, jurisdiction.
P 04
Digital
Sovereignty
The right to push back, to consent, to be represented accurately. Rests on the three above.
You cannot exercise the fourth without the first three underneath it.
The levels
Three levels at which sovereignty is held — or lost.
L 01
Micro
the individual
Your data, your image, your voice, your work. The locus most artists are fighting from.
L 02
Meso
the community
The institution, the union, the trade body. Where collective terms are negotiated. The locus SAG-AFTRA built from.
L 03
Macro
the nation
The regulator. The policy environment. Where your country stands in the global digital architecture.
The matrix
Where each case lands on the rubric.
Image likeness in training
Opt-in vs opt-out knowledge
Bartz · individual authors
Right to push back
Union member registries
SAG-AFTRA bargaining literacy
UMG · WMG · Sony licensing
12 AI provisions, 2026
EU AI Act Art. 53 disclosure
largely absent globally
NO FAKES Act · state law
no settled framework
Filled cells are where the fight is happening. Faint cells are where it isn't — yet.
In conversation with
The lineages this framework draws on and extends.
Indigenous Data Sovereignty
Carroll, Garba, Plevel et al. · CARE Principles · Te Mana Raraunga
The originating lineage for sovereignty applied to data. Extended here from Indigenous-within-settler-state to Commonwealth nation-state contexts.
Critical AI Studies
Crawford · Birhane · Noble · Bender et al.
Bias, extraction, and infrastructural critique of AI systems. Grounds the structural argument about why sovereignty is being lost in the first place.
Digital Colonialism
Couldry & Mejias · Kwet · Mhlambi
Data extraction as continuity of colonial logics. The framing of who is the source vs who is the beneficiary of the digital economy.
Digital Sovereignty (Policy)
Couture & Toupin · Floridi · Pinto · Mueller
Sovereignty as defence (control your data, protect your borders) — this framework reframes it as presence (be in the record, contribute to it).
AI in Education / Global South
Selwyn · Pinkwart · Adams Becker · Holmes
The educational substrate through which digital literacy is either built or skipped over. The site at which sovereignty at scale becomes possible.
Sesotho & Low-Resource NLP
Mokhosi et al. · Moorosi (DAIR) · ≈ParlAfrika
The empirical thinness of indigenous-language AI capability — itself a sovereignty argument. Grounds the language layer of the framework.
Where the fight is happening
Jurisdictions
with the
infrastructure.
- US courts — Bartz, NYT v. OpenAI, Disney v. Midjourney
- SAG-AFTRA — collective bargaining with the AMPTP
- EU AI Act — Art. 53 disclosure, full enforcement Aug 2026
- NY / California / Colorado / Montana — state-level replica and neural-data laws
Where it isn't — yet
Jurisdictions
without it.
- Most of the Global South — no comparable AI copyright case law in force
- Many jurisdictions — no comprehensive data protection law in force
- Non-union performers globally — no contractual AI floor
- Indigenous-language workers — absent from training, absent from terms
Where the framework points · the proposal
Existing disclosure says "this was AI." It does not say what that means.
A substantive standard distinguishes AI as finishing touch from AI as substitute capability. The threshold: if you could not have produced at least 75% of the output without AI, the audience deserves to know not just that AI was used, but how. The deeper test underneath the threshold is the resilience test — if the tool goes down, are you still here? Sovereignty over a capability means removing the tool does not remove the capability.
A three-tier disclosure regime
Capability · Impact · Precaution.
D 01
Capability
disclosure
What can this AI actually do? Producer-side, output-specific. The 75% threshold lives here.
D 02
Impact
disclosure
What might this AI do at scale? Platform-side, hypothetical. The deployment context, the population affected, the foreseeable harm.
D 03
Precautionary
regulation
What safeguards protect each layer of sovereignty? Regulator-side, structural. Maps onto Micro · Meso · Macro.
Disclosure is the operational bridge between the literacy pillar and the sovereignty pillar.
Disclosure as a platform capability
What's been built. What can be built next.
C2PA · Content Credentials
Est. 2021 · 6,000+ members
Open standard. Cryptographically signed manifest embedded at point of creation. Steering committee: Adobe, BBC, Intel, Microsoft, Google, OpenAI, Meta, Sony, Truepic, Publicis Groupe.
Google · SynthID
Unified detector · May 2025
Invisible watermarking across text, image, audio, video. Open-sourced for text via Hugging Face. Embedded in Gemini, Imagen, Veo, Lyria, NotebookLM.
IPTC 2025.1
Ratified Nov 2025
Four new XMP metadata fields specifically for AI-generated and AI-assisted content. Parallel to C2PA; together they form the regulator's reference stack.
Hardware-level provenance
Pixel 10 · Galaxy S25 · 2025
C2PA signing at the chip level (Titan M2 on Pixel 10). Native camera integration. Provenance is moving down the stack.
Native AI-written disclosure
The proposal
Every output, every export: the AI itself writes a disclosure of what it contributed. In its own voice. What inputs it had. Where the human was doing the work, where the AI was auxiliary. A default platform capability, not a user burden.
Shared responsibility
Innovators · owners · regulators
Standardisation of disclosure protocol sits with the platforms that produce the outputs as much as it sits with regulators. The platforms can make honesty available, as a feature.
Sources: C2PA Coalition · Google DeepMind · IPTC
Who deepfake law protects today
The exposed
minority.
- Public figures — right-of-publicity laws, mostly commercial-use focused
- NCII victims — TAKE IT DOWN Act, 47 US states. Sexual content only.
- Political candidates — 30 US states with election deepfake disclosure
- Danish citizens — Denmark 2025 amendment, the only law treating every person's likeness as IP by default
Who it doesn't
Everyone
else.
- Voice-cloning fraud targets — the grandparent scam, CEO impersonation calls
- Non-sexual reputation damage — defamatory deepfakes, not covered
- Identity theft and impersonation — workplace, social, financial
- Student-on-student deepfake harassment beyond sexual content
- Almost every African jurisdiction — no individual deepfake protection in force
Act 04
04
The African convergence.
Why African institutions need to build the infrastructure now — and what one such build looks like.
The argument
If sovereignty over the AI layer is being negotiated by those with the infrastructure — then African institutions need to be building that infrastructure. Not after consolidation. Now.
The global AI landscape is consolidating around a handful of model providers, language families, and infrastructure architectures. The longer that consolidation runs without African institutional presence as a counterweight, the harder it becomes to insert African languages, African pedagogies, and African sovereignty considerations into systems that are now training the next generation.
The build
GEM Africa.
A continental architecture for women's economic empowerment and AI literacy at scale.
A continental expansion of the thirteen-year-old Gender Entrepreneurship Empowerment and Media Institute, launched in convergence with Code Cafe AI. Built on the principle that digital literacy delivered at women's empowerment scale, and women's empowerment delivered with an AI literacy spine, are two adjacent problems that no longer admit being solved separately.
Year 00 · Origin
Founding country
Where GEM has thirteen years of practice. The methodology and language layer originate here.
Year 01 · Pilot
South Africa
Confirmed school partnership. Implementation context ready.
Year 02 · Expansion
Kenya
Founding country lead in advanced discussions. Full inception cycle for institutional partnership.
The methodology
The Aya Learn Model. Six phases. Indigenous-language-capable from the methodology layer, not as a translation overlay.
A pedagogical framework spanning the full arc from digital presence, through digital literacy and digital economy participation, to digital sovereignty. Phases 1–4 are open educational resources for broad ecosystem adoption. Phases 5–6 are delivered through the patented Aya learning platform as proprietary programme content. Sesotho is the first deployed language layer, drawing on parallel work on Thuto ya AI ka Sesotho.
Digital literacy at scale is the architecture that lets African communities be in the conversation at all.
The historical beat
We have inherited the terms before. With social media. With the internet itself. With the platforms that now shape how we are surfaced and represented in the digital record.
We do not have to do it again. But the window for not doing it again is narrow, and it is closing.
A question for you · at all three levels
What does digital sovereignty mean
for you?
Micro. Your data, your face, your voice, your work, your right to be represented accurately and to push back when you're not.
Meso. Your community, your profession, your institution, the collective vehicles through which you negotiate the terms of your participation.
Macro. Your country, your region, the position your government holds, or does not hold, in the global digital architecture.
Maletheba Lesa Makoele
June 2026