Humanein the Loop
← The matrix
07

AI power should be balanced in society

Current path

AI power concentrates in a handful of firms with unprecedented influence over markets, public discourse, and policy. Open alternatives are starved of capital and compute. Democratic institutions struggle to hold AI firms accountable.

Better future

AI power is distributed across a healthy ecosystem of commercial, open-source, public, and cooperative actors. Antitrust and competition policy apply. Democratic institutions retain capacity to govern.

Drift across the three domains

Norms

Mixed3 signals
CHT recommends
  • Treat AI market concentration as a competition and democratic concern.
  • Sustain substantive public debate on open vs closed AI beyond either-side capture.
Indicators we track
  • 7.N.aAntitrust and competition discourse applied to AI
  • 7.N.bOpen-source vs closed-source debate
  • 7.N.cConcerns about democratic capture

Laws

Advancing7 signals
CHT recommends
  • Apply antitrust and competition rules to AI markets.
  • Mandate interoperability and data portability.
  • Fund public-option AI and sovereign compute alternatives.
  • Restrict political use of AI (deepfakes, AI-generated campaign speech).
Indicators we track
  • 7.L.aAntitrust action against AI market concentration
  • 7.L.bInteroperability and data portability mandates
  • 7.L.cPublic-option AI and sovereign compute funding
  • 7.L.dRestrictions on political uses of AI

Design

·
Insufficient data1 signal
CHT recommends
  • Support responsible open-weights releases of capable models.
  • Invest in decentralized and federated architectures.
  • Provide fair API access to closed models for third-party developers.
Indicators we track
  • 7.D.aOpen-source model releases
  • 7.D.bDecentralized and federated architectures
  • 7.D.cThird-party access to closed models

Recent signals

AdvancingNorms · 7.N.cEUROPEApr 18, 2026

Data center lobby secured EU provision to keep environmental impact data confidential - report

A new report reveals that a lobbying group representing major tech companies, including Microsoft, Amazon, Google, and Meta, successfully secured a provision in the EU to keep data center environmental impact data confidential.

WhyA new report scrutinizes tech giants' lobbying efforts, surfacing their influence in securing an EU provision to hide environmental data.Concerns about democratic capture
RegressingLaws · 7.L.bUSApr 9, 2026

Comparison Shopping Is Not a (Computer) Crime

A federal district court ruled that Perplexity's AI-enabled browser violated the Computer Fraud and Abuse Act by accessing Amazon's website to help users comparison shop. The EFF has filed an amicus brief supporting Perplexity's appeal to the Ninth Circuit, arguing the decision harms competition and innovation.

WhyA federal court ruled Perplexity's AI shopping agent violated the CFAA by accessing Amazon, legally protecting a walled garden.Interoperability and data portability mandates
AdvancingDesign · 7.D.cGLOBALApr 1, 2026

Anthropic Fellows Program opens next two cohorts (May & July 2026)

Anthropic opened applications for its next Fellows cohorts beginning May and July 2026. 4-month program, $3,850/week stipend + ~$15k/mo compute funding + mentorship for independent safety researchers. 40% of first-cohort fellows joined Anthropic full-time; 80%+ produced papers.

WhyThird-party access channel widens; structured external-researcher pipeline at frontier lab, though scale remains small (dozensThird-party access to closed models
AdvancingLaws · 7.L.cEUROPEMar 30, 2026

Mistral raises €722M debt to build sovereign European AI compute

Mistral AI secured €722M ($830M) in debt financing from seven-bank consortium led by Bpifrance and BNP Paribas to build a 44MW data center near Paris with 13,800 Nvidia GB300 GPUs (Q2 2026 target). Largest AI-focused debt raise by a European technology company to date; targets 200MW capacity by end-2027.

WhyEuropean champion scales outside US hyperscaler dependence; credible sovereign alternative emerging on the compute layerPublic-option AI and sovereign compute funding
MixedLaws · 7.L.bEUROPEJan 8, 2026

EC publishes DMA review summary with AI-gatekeeper scope under active consideration

On Jan 8, 2026 the Commission published the summary of 450+ contributions to its DMA review consultation (ran Jul-Sep 2025, with dedicated AI questionnaire launched Aug 26, 2025). AI emerged as the most prevalent theme; respondents split on whether to add a standalone AI Core Platform Service category. Article 53 report due May 3, 2026.

WhyConsultation-stage move; norm-gathering for future enforcement rather than action.Interoperability and data portability mandates
AdvancingLaws · 7.L.dUSJan 1, 2026

Nevada AI political advertising disclosure law (AB73) takes effect

Nevada AB73 became effective Jan 1, 2026, requiring "clear and conspicuous" disclosure when synthetic/AI-generated media is used in political communications. Unanimously passed; gives depicted candidates injunctive relief against undisclosed AI manipulation. Part of a broader wave of state deepfake-political laws.

WhyState-level democratic-integrity guardrails spreading; counterweight to federal inaction.Restrictions on political uses of AI
AdvancingMajorLaws · 7.L.cEUROPEDec 5, 2025

Council of EU adopts conclusions on European competitiveness in the digital decade

On Dec 5, 2025 the Council published conclusions on European Competitiveness in the Digital Decade, urging open standards, interoperability and reduced vendor lock-in in cloud, AI, cybersecurity and connectivity; asks Commission to develop common criteria for sovereign cloud services ahead of the forthcoming Cloud and AI Development Act (Commission proposal due Q1 2026).

WhyCouncil-level political backing for sovereign EU compute and public AI infrastructure ahead of CADAPublic-option AI and sovereign compute funding
RegressingNorms · 7.N.cUSNov 17, 2025

House GOP eyes NDAA as vehicle to revive AI state-law preemption

On Nov 17, 2025, House Majority Leader Scalise confirmed to Punchbowl News that House GOP leaders are exploring attaching AI preemption language to the FY26 National Defense Authorization Act. Congressional Progressive Caucus and states-rights Republicans lined up against.

WhyIndustry allies retry democratic capture via a different legislative vehicle after the 99-1 defeatConcerns about democratic capture
AdvancingMajorDesign · 7.D.aGLOBALSep 29, 2025

DeepSeek V3.2-Exp ships Sparse Attention with 50%+ API price cut

DeepSeek released V3.2-Exp on Sep 29, 2025 under MIT license, debuting DeepSeek Sparse Attention (DSA) for long-context efficiency and cutting API prices 50%+ immediately. Full V3.2 shipped Dec 1, 2025 (with V3.2-Speciale reasoning variant).

WhyCost collapse continues; compute-access moat at the frontier eroding through architectural innovation, not just scale.Open-source model releases
AdvancingMajorLaws · 7.L.aUSSep 12, 2025

FTC-driven Microsoft-OpenAI MOU restructures $13B partnership

In September 2025 Microsoft and OpenAI signed a non-binding MOU to restructure their nearly $13B relationship, enabling OpenAI to transition toward a for-profit entity with freedom to partner with rival cloud providers. The restructuring was driven by threat of a formal FTC merger challenge treating multi-billion-dollar exclusive licensing as an undisclosed merger.

WhyFirst concrete antitrust-driven unwinding of a cloud-AI quasi-merger. Sets precedent that exclusive frontier-lab/cloud partnerships can beAntitrust action against AI market concentration
AdvancingMajorDesign · 7.D.aGLOBALAug 5, 2025

OpenAI releases gpt-oss-120b and gpt-oss-20b under Apache 2.0

First OpenAI open-weight release since GPT-2. 117B and 21B total-parameter MoE reasoning models (5.1B and 3.6B active), Apache 2.0 license, near-parity with o4-mini / o3-mini on core benchmarks. Runs on single 80GB GPU / 16GB consumer hardware respectively.

WhyFrontier-lab capitulation to open-weights norm post-DeepSeek R1; design-layer power diffusion acceleratingOpen-source model releases
AdvancingMajorNorms · 7.N.cUSJul 1, 2025

US Senate strips AI state-law moratorium from OBBBA 99-1

Senate voted 99-1 during vote-a-rama to strike the Blackburn/Cantwell (bipartisan) amendment removing the 10-year (later 5-year) federal preemption of state AI laws from the One Big Beautiful Bill Act. Only Sen. Tillis (R-NC) voted no.

WhyNear-unanimous defeat of industry-backed preemption; preserves plural regulatory venues against capture and protects ~1,000 pending stateConcerns about democratic capture