The Elephant in the AI Room

If you’ve followed my recent posts, you’ve seen we’re in a brutal transition:

  • 📊 Value migrated from execution to strategy
  • 💰 The SaaS model is under existential threat
  • 🧠 AIs have clear limits in reasoning
  • 🛣️ Professionals must choose one of 3 roads

But what happens behind the scenes at the giants controlling this intelligence?

February 2026 brought a controversy mixing:

  • 🕵️ Industrial espionage
  • 🛡️ National security
  • 😏 A good dose of irony

The Case of “Forbidden Distillation”

What Happened

Anthropic (creators of Claude) identified an industrial-scale campaign from three Chinese labs:

  • 🔴 DeepSeek
  • 🔴 Moonshot
  • 🔴 MiniMax

The Staggering Numbers

The attack:

  • 📧 24,000 fake accounts created
  • 💬 16 million conversations generated
  • 🎯 Goal: Extract Claude’s advanced capabilities
  • ⏱️ Period: Months of coordinated operation

The Technique: “Knowledge Distillation”

How it works:

Traditional Method (Expensive):
→ Collect billions of texts
→ Build infrastructure ($100M+)
→ Train for months ($50M+ in GPUs)
→ Iterate and refine
= Total cost: $200M - $500M
= Time: 12-18 months

Distillation Method (Cheap):
→ Create thousands of fake accounts
→ Bombard superior AI with questions
→ Collect all responses
→ Train small model with these responses
= Total cost: $5M - $20M
= Time: 2-4 months

The math is brutal:

  • 95% cheaper
  • 80% faster
  • Similar result (not identical, but close)

The Distillation Process in Detail

Step 1: Mass Account Creation

Automated script creates:
- 24,000 fake Gmail accounts
- All subscribe to Claude Pro ($20/month)
- Initial cost: ~$480k/month

Step 2: Question Bombardment

# Conceptual example
for topic in knowledge_domains:
    for complexity in [easy, medium, hard]:
        question = generate_question(topic, complexity)
        response = claude.ask(question)
        database.save(question, response)

Result: 16 million high-quality question-answer pairs.

Step 3: Training Own Model

Data: 16M conversations with Claude
Base model: Llama 3 or similar (open-source)
Fine-tuning: Using Claude's responses as "truth"
Result: Model that mimics 70-80% of Claude

Absurd cost-benefit:

  • Anthropic spent: ~$500M+ developing Claude
  • DeepSeek spent: ~$15M copying

The Reaction: National Security

The Official Narrative

Anthropic and OpenAI argue:

❌ “This is industrial espionage” ❌ “Threat to US national security” ❌ “Intellectual property theft” ❌ “Unfair advantage for China”

Measures Taken

Immediate actions:

1. Blocking 24,000 accounts
2. IP usage limits
3. Automated pattern detection
4. Stricter identity verification
5. Aggressive rate limiting

Political actions:

  • 🏛️ US Congress lobbying
  • 📋 Regulation proposals
  • 🚫 Possible blocking of Chinese API access
  • 💼 Discussions about “sovereign AI”

The Great Market Irony

This is where the story gets interesting (and hypocritical).

The Spell Backfired

Crucial question:

How can companies that scraped the entire internet without permission to train their models complain that others are “stealing” their work?

The Exposed Hypocrisy

How OpenAI, Anthropic and Google trained their models:

Data used WITHOUT authorization:
✗ Books from LibGen and Z-Library (piracy)
✗ GitHub code (licenses ignored)
✗ News site articles (no compensation)
✗ Reddit and Twitter posts (no consent)
✗ Art from millions of artists
✗ Songs and lyrics (copyrights?)
✗ YouTube video transcriptions
✗ Academic PDF documents

Conservative estimate:

  • 📚 10+ trillion words
  • 🎨 5+ billion images
  • 💻 100+ million code repositories
  • 🎵 Millions of creative works

Estimated value if they had PAID for rights:

  • 💰 $50 billion - $200 billion

What they actually paid:

  • 💸 $0

The Irony Equation

Anthropic training Claude:
"Let's use everything from internet without asking!"
→ OK, it's 'fair use' for research

DeepSeek distilling Claude:
"Let's use Claude without asking!"
→ CRIME! ESPIONAGE! NATIONAL SECURITY!

See the irony?

The Debate Voices

🔴 Position 1: It IS Theft

Arguments:

  • Claude is Anthropic’s intellectual property
  • Cost hundreds of millions to develop
  • Terms of service prohibit this use
  • Unfair competitive threat

🟢 Position 2: Spell Backfired on Wizard

Arguments:

  • Anthropic did the same to human creators
  • Didn’t compensate millions of authors/artists
  • “Fair use” applies to some but not others?
  • Total hypocrisy

🟡 Position 3: Both Are Wrong

Arguments:

  • Nobody should be able to copy without permission
  • But Anthropic has no moral high ground
  • Entire system is flawed
  • We need new laws

The Geopolitical Context

The USA vs China Tech War

Behind the controversy, there’s a technological cold war:

USA:

  • 🏆 Dominates cutting-edge AI (for now)
  • 💰 Massive investment ($100B+)
  • 🛡️ Controls access via export controls
  • 🎯 Strategy: Maintain technological leadership

China:

  • 🎯 Goal: Lead AI by 2030
  • 💰 Massive state investment
  • 🔓 Strategy: Copy + innovate + scale
  • ⚡ Advantage: Execution speed

The Stakes Are Extremely High

Whoever dominates AI will dominate:

  • 🪖 Military supremacy
  • 💵 Global economy
  • 📊 Surveillance and control
  • 🧬 Scientific research
  • 🎓 Education
  • 🏭 Industrial production

This isn’t about “software theft.” It’s about geopolitical power in the 21st century.

And the End User? (Us)

Benefits of the “War”

More AI options

  • DeepSeek offers alternative to Claude
  • Competition forces innovation
  • Prices drop (some Chinese models are free)

Democratization

  • Powerful models reach more countries
  • Entry barrier decreases
  • Open source strengthens

Damages from the “War”

Fragmentation

  • Internet “balkanized” by countries
  • Chinese can’t use Claude
  • Westerners may be blocked from DeepSeek
  • Less global collaboration

Less transparency

  • Companies hide more information
  • “Security” becomes excuse for secrecy
  • Users know less about what they use

Security risks

  • AI arms race
  • Less time for safety research
  • Rush can cause accidents

Back to the 3 Roads

With all this war happening at the top, how should you position yourself?

🎯 Road 1: The Orchestrator

New reality in 2026:

  • AI agents have full computer access (sandboxes)
  • Execute technical tasks end-to-end
  • Your role: Conduct the orchestra, not play instruments

War impact:

  • More tools available (USA + China)
  • Must master multiple platforms
  • Arbitrage between Claude, GPT, DeepSeek

⚙️ Road 2: The Systems Builder

New reality:

  • Infrastructure for agents to operate safely
  • Cost optimization (right model at right time)
  • Smart routing layers

War impact:

  • More complexity (different APIs, restrictions)
  • Opportunity (companies need multi-model systems)
  • Security becomes critical

🔬 Road 3: The Domain Translator

New reality:

  • Specialist (law, medicine, engineering)
  • Uses AI to solve real niche problems
  • Deep context is differentiator

War impact:

  • Less affected (war is between giants)
  • Can use tools from any origin
  • Focus on problem, not technology

This road may be safest during tech war.

The Uncomfortable Truth

The AI Paradox in 2026

AI is trying to self-improve (RSI), but:

❌ Still “cheats” on performance tests ❌ Fails at novel mathematical problems ❌ Can’t do pure reasoning (only pattern recognition) ❌ Built on “stolen” data ❌ Controlled by few countries/companies ❌ Used as geopolitical weapon

But even so:

✅ Completely changing work ✅ Destroying business models ✅ Creating value (and destroying old value) ✅ Forcing everyone to adapt

Where Is the “Gold” of 2026?

NOT in:

  • ❌ Code (became commodity)
  • ❌ SaaS software (dying as “per seat”)
  • ❌ Technical execution (AI does this)
  • ❌ Having access to best AI (access war)

It’s in:

  • Strategic direction (what to build?)
  • Deep context (understanding the problem)
  • Orchestration (using AI as tool)
  • Adaptability (world changes fast)
  • Platform independence (don’t depend on one AI)

Conclusion: Living in the Age of Hypocrisy

We’re in a fascinating and disturbing moment:

Companies that:

  • Built empires by copying without permission…
  • Now cry when they’re copied
  • And use “national security” as shield

Meanwhile:

  • Users caught in the middle of war
  • Original creators still not compensated
  • Tech race accelerates without brakes
  • Ethics thrown out window

The question isn’t if this is right or wrong.

The question is: How will you navigate this chaos?


Reflection

Do you think AI labs have the moral high ground to complain about data “theft”?

Or are we seeing the “spell backfire on the wizard”?

Does this tech war benefit or harm the end user?

And more importantly: How are you positioning yourself in this scenario?


Your Controversial Opinion is Welcome

This is the kind of debate with no easy answers.

Share your vision (even if controversial):

The future is being decided now.

Will you watch or participate?


Read Also