AI Ethics · Wzdom Research · Q2 2026

AI affects your job, credit,
and health. Who's accountable?

The same AI systems companies use to screen job applications, approve loans, and triage healthcare are operating without meaningful oversight in most of the world. This guide covers what the law actually requires, what companies have genuinely committed to, and — most importantly — what you can do when they fall short.

Published: May 22, 2026 Edition: Q2 2026 Author: Wzdom Research Regulations tracked: 12 Companies reviewed: 10 Read time: ~10 min

The law is still catching up. Wzdom Privacy monitors what actually leaves your device, right now.

Download free →
12
Jurisdictions tracked
EU, US, UK, China, Canada, Brazil + 6 more
4
EU AI Act risk tiers
Unacceptable → High → Limited → Minimal
10
AI companies reviewed
OpenAI, Google, Meta, Microsoft, Anthropic + 5
Jump to law tracker ↓

While regulators debate — your data moves

Laws take years to pass. Wzdom Privacy monitors what actually leaves your device, right now.

100% on-device · No account required · Free during beta

Download free

Why this matters right now

AI is making decisions about your life — without your knowledge.

AI systems are not just chatbots. The same technology is being used to screen your job application, set your insurance premium, flag your loan, and determine your bail. Most people never find out. Most jurisdictions have no rules requiring that they be told.

Hiring & recruitmentCV screening, automated interview scoring, background check classification — often with no human review for rejection.
Insurance & creditRisk scoring models use data you never consented to share. You may be denied without knowing an AI was involved.
Healthcare & medical triageDiagnostic AI tools are influencing treatment priority and resource allocation in hospitals across the EU and US.
Law enforcementPredictive policing and facial recognition are deployed in cities with minimal oversight and near-zero public disclosure.
Content & informationWhat you see, read, and believe is algorithmically curated. Recommendation systems are the most powerful editorial force in history.
Financial decisionsAlgorithmic trading, fraud detection, and credit scoring systems affect billions of transactions with no meaningful appeal mechanism.

In most cases, you will never know what system made the decision, what data it used, whether bias affected the outcome, or who can be held accountable. That is the problem Wzdom Research tracks.

Core principles

The 7 core AI ethics principles

There is no global AI ethics standard — yet. But there is broad consensus across the EU AI Act, OECD AI Principles, IEEE Ethically Aligned Design, and most corporate ethics frameworks around seven foundational ideas. These principles are the lens through which we evaluate every AI system.

Fairness & Non-discrimination

AI systems must not produce discriminatory outcomes based on race, gender, age, disability, or other protected characteristics. Bias in training data propagates into real-world harm.

Transparency

Users must be told when they are interacting with AI, and must have access to meaningful explanations of how AI decisions that affect them are made.

Safety & Security

AI systems must be robust against adversarial attacks, misuse, and failure. The consequences of failure must be proportionate to the degree of human oversight maintained.

Privacy

AI must respect data protection principles: data minimisation, purpose limitation, and the right to erasure. Training data must be lawfully obtained.

Human Oversight

High-stakes AI decisions — hiring, credit, healthcare, law enforcement — must have meaningful human review. Fully automated consequential decisions require explicit legal authority.

Environmental Responsibility

Training large AI models consumes enormous energy. AI companies must disclose carbon footprint and commit to measurable reduction targets.

Accountability

There must always be a legal person (company or individual) accountable for an AI system's outcomes. "The algorithm decided" is not an acceptable legal or ethical defense.

Wzdom's position

What we believe about AI

Wzdom Research doesn't just report on AI ethics — we have a position. These are the principles we apply to our own products and hold other AI companies to. We publish them so you can hold us to the same standard.

  • AI systems that affect people's lives must be transparent — users must know when AI is involved in a decision.
  • AI must be privacy-respecting by default — data minimisation is not a compliance checkbox, it is a design requirement.
  • High-stakes AI decisions must be explainable and reviewable — "the algorithm decided" is never an acceptable final answer.
  • There must always be a human accountable for an AI system's outcomes — accountability cannot be delegated to a model.
  • AI should preserve user autonomy — systems that manipulate, exploit, or surveil users without meaningful consent are unethical regardless of legal status.
  • AI that runs entirely on your device is the highest-trust architecture possible — it is the model Wzdom builds to.

Global regulation tracker

Where AI regulation stands — worldwide

12 jurisdictions tracked across Europe, the Americas, Asia-Pacific, the Middle East, and South Asia. Hover any highlighted country for details.

Loading map…
Binding law
Active — voluntary
Active — state controlled
Pending / partial
No tracked framework
🇪🇺
European Union
EU AI Act (2024)
Binding Law
StatusIn force — phased enforcement 2025–27
Enforcement€30M or 6% global revenue
🇺🇸
United States
EO 14179 (2025) + FTC §5
Active — Voluntary
StatusSoft / deregulatory; no federal AI law
EnforcementFTC civil fines (case by case)
🇬🇧
United Kingdom
AI Security Institute + Opportunities Plan
Active — Voluntary
StatusPro-innovation; sector regulators lead
EnforcementICO, FCA, Ofcom (sector-by-sector)
🇨🇦
Canada
Directive on AI + AIDA (C-27 lapsed)
Pending
StatusC-27 lapsed Jan 2025; new bill expected
EnforcementCAD $25M or 3% revenue (proposed)
🇨🇳
China
GenAI Regulations (2023) + Algorithm Rules
State Controlled
StatusLicence required; state content oversight
EnforcementMIIT enforcement; licence revocation
🇯🇵
Japan
AI Guidelines for Business 2024 (METI)
Active — Voluntary
StatusG7 Hiroshima AI Process signatory
EnforcementAgency guidance; no binding AI law
🇰🇷
South Korea
AI Framework Act (Jan 2025)
Binding Law
StatusFirst binding AI law in Asia-Pacific
EnforcementMinistry of Science & ICT
🇦🇺
Australia
Voluntary AI Safety Standard 2024
Active — Voluntary
StatusMandatory guardrails under consultation
EnforcementDepartment of Industry guidance
🇸🇬
Singapore
Model AI Governance Framework + AI Verify
Active — Voluntary
StatusWorld-leading governance toolkit
EnforcementIMDA guidance; no binding AI law
🇦🇪
United Arab Emirates
National AI Strategy 2031 + TDRA Guidelines
Active — Voluntary
StatusStrategy-led; positioning as global AI hub
EnforcementTelecom & Digital Gov. Regulatory Auth.
🇸🇦
Saudi Arabia
National AI Strategy 2030 (SDAIA)
Active — Voluntary
StatusSDAIA leads national AI governance
EnforcementSaudi Data & AI Authority (SDAIA)
🇮🇳
India
Digital Personal Data Protection Act 2023
Pending
StatusDPDP Act passed; rules still pending
EnforcementData Protection Board of India (forming)

Global regulations

Global AI ethics regulations — 2026 status

AI ethics is finally moving from voluntary commitments to binding law — but the pace is slower than the technology. The EU has led with the world's first comprehensive AI regulation, and the penalties are real. The US lags significantly — no federal law exists, and the patchwork of executive orders and FTC enforcement leaves most people unprotected. Here's exactly where every major jurisdiction stands, and what it means for you.

Jurisdiction Framework Status Ethics requirements Enforcement
European Union EU AI Act (2024) Active Transparency, human oversight, bias testing, right to explanation — mandatory for high-risk AI €30M or 6% global revenue (max)
United States EO 14179 (2025) + FTC §5 Active (soft) Biden's EO 14110 (safety mandates) revoked Jan 2025. Trump's EO 14179 removes AI restrictions — no mandatory ethics requirements. FTC applies existing consumer protection law. FTC deceptive practices fines (civil)
United Kingdom AI Security Institute + Pro-Innovation Approach Active Voluntary principles; AI Security Institute (renamed from AI Safety Institute, Feb 2025) evaluates frontier models. Sector-specific rules via ICO, FCA, Ofcom. Sector regulators (ICO, FCA, Ofcom)
Canada Directive on AI + AIDA (Bill C-27 lapsed) Pending Bill C-27 (AIDA) lapsed when parliament was prorogued Jan 2025 — would have required explainability and human oversight for high-impact AI. New government may reintroduce. CAD $25M or 3% global revenue (proposed)
China Generative AI Regulations (2023) Active Content labelling, no illegal discrimination, CCP-aligned "socialist core values" MIIT enforcement; license required
OECD OECD AI Principles (2019, updated 2024) Active (voluntary) Inclusive growth, human-centered values, transparency, robustness, accountability None — peer pressure and domestic adoption
UNESCO Recommendation on AI Ethics (2021) Active (voluntary) Human rights framing, environmental responsibility, gender equality, data governance None — member state reporting
G7 / Hiroshima AI Process Hiroshima AI Process Code of Conduct (2023) Active (voluntary) 11 guiding principles: transparency, identify AI-generated content, report incidents None — voluntary industry commitments

"Every major AI company has published AI ethics principles. Most are non-binding, unverifiable, and written by the same organizations they're supposed to constrain. That's not ethics — that's marketing."

— Wzdom Research · AI Ethics Index Q2 2026

EU AI Act

EU AI Act — the 4-tier risk model

The EU AI Act classifies AI systems by risk level. The higher the risk, the stricter the requirements. Most consumer AI tools fall in the Limited or Minimal tiers. AI used in hiring, credit, medical diagnosis, or law enforcement is High-risk and subject to mandatory conformity assessment, logging, and human oversight.

Tier 1

Unacceptable Risk

Banned outright in the EU.

  • Real-time biometric surveillance
  • Social credit scoring
  • Subliminal manipulation
  • AI that exploits vulnerable groups
Tier 2

High Risk

Mandatory conformity assessment, logging, human oversight.

  • AI in hiring/recruitment
  • Credit scoring
  • Medical device AI
  • Law enforcement AI
  • Education admissions
Tier 3

Limited Risk

Transparency obligations only — must tell users they're talking to AI.

  • Chatbots & AI assistants
  • Deepfake content
  • AI-generated text disclosure
Tier 4

Minimal Risk

No mandatory requirements. Voluntary codes of conduct encouraged.

  • AI-powered spam filters
  • Recommendation systems
  • AI in video games

Company commitments

How AI companies' ethics commitments actually measure up

Every major AI company has published AI ethics principles. We read them all. Most are non-binding, unverifiable, and written by the same organizations they're supposed to constrain. The most common tell: a company's ethics page is updated more frequently than its safety incident disclosures. Below is an honest assessment of where commitments appear genuine — and where the gap between stated values and actual practice is widest.

Anthropic
Claude · claude.ai
Leading
  • + Constitutional AI — embedded ethical constraints in training
  • + Responsible Scaling Policy with explicit capability thresholds
  • + Published safety research — unusually transparent for a lab
  • ~ Safety Board dissolved and reconstituted under pressure (2024)
OpenAI
ChatGPT · GPT-4
Moderate
  • + Red-teaming published pre-launch for GPT-4
  • + Usage policy prohibits weapons, CSAM, deception
  • Safety team resignations raised governance concerns (2024)
  • Mission drift: "for humanity" vs commercial pressure
Google / DeepMind
Gemini · Bard
Moderate
  • + Published "Responsible AI practices" with concrete guidelines
  • + PAIR (People + AI Research) team does public-facing work
  • Fired Ethical AI team leaders in 2020 amid controversy
  • AI in military contracting raised ethics team objections
Microsoft
Copilot · Azure AI
Moderate
  • + Responsible AI Standard — one of most detailed published
  • + Office of Responsible AI with executive accountability
  • Recall (Windows AI screenshot feature) bypassed ethics review
  • ~ Large military AI contracts with limited public disclosure
Meta
Meta AI · Llama
Lagging
  • + Open-weights Llama releases support external safety research
  • Systemic algorithmic harm documented (Frances Haugen, 2021)
  • No independent AI ethics board with real authority
  • EU blocked Meta AI rollout over privacy concerns (2024)
xAI (Grok)
Grok · X
Lagging
  • + Open-weight Grok model releases
  • No published ethics framework or safety report
  • CEO public statements undermine AI safety field
  • EU Privacy regulator investigation active (2024)
Mistral AI
Le Chat · Mistral 7B
Leading
  • + Open-weight models enable independent safety review
  • + EU-based — subject to AI Act from day one
  • + No advertising revenue eliminates key misalignment
  • ~ Smaller team limits dedicated ethics research capacity
DeepSeek
DeepSeek R1/V3
Lagging
  • + Open-weight models release for research
  • China jurisdiction — mandatory CCP content controls
  • No independent ethics board or published safety framework
  • Data handling opaque; US ban discussions ongoing

Research methodology

How we evaluate AI ethics commitments

Every company assessment, regulation summary, and stance rating in this guide is based on independently verifiable sources. We do not accept briefings from AI companies as the basis for positive ratings — everything must be externally confirmable.

  • Publicly available privacy policies, terms of service, and ethics frameworks — as published by the companies themselves
  • Regulatory filings, enforcement actions, and official guidance from EU DPAs, the FTC, ICO, and other authorities
  • Academic research, peer-reviewed analysis, and independent audit findings
  • Documented incidents — safety team resignations, whistleblower reports, regulator investigations — with original source links
  • Verified updates — company stances are reviewed on a rolling basis as policies change, new incidents emerge, or regulatory status shifts

Findings may be cited with attribution to Wzdom Research. Each entry in the developments feed includes a primary source link. We do not take advertising from or accept payment from any company reviewed in this guide.

📄
Wzdom Research — Full Paper

Privacy in the Age of Frontier AI: a scored analysis of 12 AI tools across six privacy dimensions, with regulatory timeline, scoring rubric, and primary citations.

Recent developments

AI ethics developments — 2026

The AI ethics landscape moves fast. These are the most significant developments from the last few months, tracked by Wzdom Research.

May 2026
European Commission publishes draft guidelines on high-risk AI classification. The Commission released draft guidance under Article 6 of the EU AI Act on how to determine whether a system qualifies as high-risk — open for stakeholder consultation until June 2026. Conformity assessment requirements take effect August 2, 2026. Source ↗
Regulation
May 2025
Irish DPC concludes engagement with Meta over AI training data. The Data Protection Commission confirmed Meta's plan to train its LLM on public EU/EEA posts could proceed after Meta implemented GDPR compliance measures — updated legitimate interest assessments, in-app objection mechanisms, and de-identification filters. Source ↗
Enforcement
Ongoing
FTC applies existing consumer protection law to AI systems. The Federal Trade Commission has stated that existing FTC Act, FCRA, and ECOA requirements apply fully to AI — covering deceptive claims, discriminatory outputs, and unfair data practices. No AI-specific exemption exists under current enforcement posture. Source ↗
Regulation
Since 2023
OpenAI enterprise and API data is not used for training by default. Data from ChatGPT Enterprise, ChatGPT Business, and the API platform has been excluded from model training by default since March 2023 — customers must explicitly opt in. This remains OpenAI's standing enterprise privacy commitment. Source ↗
Corporate
Dec 2025
UK AI Security Institute publishes Frontier AI Trends Report. AISI (renamed from AI Safety Institute in February 2025) released its inaugural evaluation of frontier AI models covering cyber capabilities, biological risk, safeguards, and software engineering — the first government-led independent evaluation programme of commercial AI at scale. Source ↗
Research

What you can do

How to hold AI companies accountable

Voluntary ethics principles are worth very little without enforcement mechanisms. Here's what actually moves companies — and what you, as an individual or researcher, can do to add meaningful pressure.

📝

File regulatory complaints

GDPR violations can be reported to your national Data Protection Authority (DPA). In the US, file FTC complaints at ftc.gov/complaint. Both have real investigative power — noyb.eu has forced €billions in fines using exactly this mechanism.

💼

Exercise data rights

Under GDPR / CCPA: request what data a company holds about you, request deletion, and opt out of automated decision-making. Most AI companies have a "Privacy Center" with these forms. Companies have 30 days to respond legally.

🗳️

Engage in consultation processes

The EU AI Act's implementing regulations are still being written. The FTC regularly takes public comment. Contact your elected representatives about AI regulation. Policy windows are open right now — this is when influence is highest.

📢

Amplify independent research

The AI ethics researchers who created real change — Timnit Gebru, Joy Buolamwini, Shoshana Zuboff — did it by producing rigorous public research that journalists and regulators could not ignore. Independent audits carry far more weight than company self-reports.

💰

Vote with your subscription

If a company's ethics practices are unacceptable, cancel the subscription. Enterprise procurement teams switching vendors over ethics concerns creates far stronger incentives than almost any single regulatory action.

🔬

Support independent AI safety labs

Organizations like the AI Security Institute (UK, formerly AI Safety Institute), METR (Model Evaluation and Threat Research), and ARC Evals do the independent evaluations that company red-teams cannot. Their findings drive regulatory action.

Privacy & ethics

Why AI ethics and AI privacy are inseparable

Privacy and ethics are not separate issues. Every AI ethics failure has a privacy dimension. Discriminatory AI requires personal data to discriminate with. Manipulative AI requires behavioral data to manipulate with. Mass surveillance AI is by definition a privacy violation. The fight for AI accountability begins with controlling what data AI systems can access about you.

Bias requires data

AI bias in hiring, credit, and healthcare is only possible because these systems were trained on, and continue to process, enormous amounts of personal data without adequate safeguards.

Manipulation requires profiles

AI systems optimized for engagement — recommenders, social media, personalised advertising — require detailed personal profiles to work. The manipulation is built into the business model.

Data minimisation = harm reduction

If an AI system cannot access your personal data, it cannot discriminate against you, manipulate you, or expose you in a breach. The best AI ethics intervention is often the simplest: collect less.

Understand the law.
Then monitor reality.

AI regulation defines what companies should do. Wzdom Privacy shows what they're actually doing — every connection your apps make, classified in plain English, in real time. On your device. Nothing leaves it.

Wzdom Privacy

You know the laws.
Now see the reality.

AI regulation tells companies what they should do. Wzdom Privacy shows you what they're actually doing — every connection your apps make, in plain English, in real time.

Download free →

macOS & Windows · Free during beta · No account needed