Limited access to cutting-edge AI: Soaring computing costs and security thresholds are changing the rules of the AI game
With Anthropic restricting the Mythos security model to white hat only, training costs topping a billion dollars, and the U.S. government accelerating its involvement — cutting-edge AI that was once open is becoming scarce and selective. Anton Leicht provides an in-depth analysis of how three major trends are changing the rules of the game in the AI industry.
Core conclusion
The "open era" of cutting-edge AI is coming to an end. Affected by the triple trend of soaring computing costs, intensifying security threats, and tightening government regulations, top AI models will become increasingly scarce and selectively available. This directly affects every AI practitioner—whether it is developers who use Claude Code to write code, or operators who use n8n to build automated workflows.
Key Points
- Incident: Anthropic’s Mythos network security model is only available to white-hat security researchers and is not available to the public
- Trend: The computational cost of training cutting-edge AI models is increasing exponentially and has exceeded the hundreds of millions of dollars threshold
- Impact: Access to AI tools in the future will be more fragmented — some models are free and open source, some are enterprise-only, and some are only open to specific groups
Background: The turning point from openness to selective openness
In April 2026, Anthropic announced the development of Mythos — a leading cybersecurity model capable of autonomously discovering and patching known vulnerabilities. But Anthropic has chosen not to make it available to the public, limiting its use to vetted white-hat security researchers. This is a landmark example of selective openness in cutting-edge AI.
Substack author Anton Leicht provides an in-depth analysis of this trend in the article "Cut Off". He pointed out that the era of "enough AI models alone to sustain a revolution" is ending** - future AI capabilities will no longer be an equal, open resource.
Key impact: Three major trends accelerate closure
| Dimensions | Changes | Implications for AI practitioners | Recommended actions |
|---|---|---|---|
| Computing costs | The cost of training cutting-edge models has exceeded hundreds of millions of dollars and is accelerating | Independent developers cannot replicate the capabilities of top models | Embrace open source models (such as DeepSeek V4) and bridge the single model gap through tool chain combination |
| Security model | Mythos proves that AI security capabilities may be used as weapons and distribution must be restricted | Some AI capabilities (especially security categories) will no longer be publicly available | Prioritize building a multi-layer security strategy and do not rely too much on the closed capabilities of a single model |
| Government supervision | Governments in the United States and other countries have tightened controls on cutting-edge AI training and deployment | Cross-border use of AI tools may be subject to new legal restrictions | Focus on local deployment solutions: LM Studio, GGUF format model, privatized inference |
Adaptation suggestions: How to deal with the era of AI fragmentation
Facing the upcoming selective openness of AI, individual practitioners and teams need to adjust their strategies:
- Embrace local and open source models — Open source models like DeepSeek V4, Mistral, Llama, etc. are still available, and can already run on consumer devices through distillation technology (such as Needle’s 26M parameter tool to call the model)
- Build toolchain combination — The less AI capabilities are encapsulated, the more toolchains are needed to make up for it. For example, the combination of Claude Code + n8n + OpenClaw can still complete end-to-end automation autonomously
- Focus on Observability and Quality Gates — Quality gates and validation processes become more important when AI output no longer comes from the "best model"
- Private deployment in advance — For key business scenarios, use the GGUF format to run the model locally without being restricted by cloud APIs
Action List
- Assess your AI workflow's dependence on specific cloud APIs
- Explore local inference options (LM Studio, llama.cpp, Ollama)
- Add quality gates and validation processes for AI output
- Pay attention to the impact of policy changes on the cross-border use of AI tools
Related extended information
- HN — 148 points hot discussion
- Cut Off — Threading the Needle
Tool entry
Already mentioned naturally in the text: Claude Code, Anthropic, DeepSeek, n8n, OpenClaw, LM Studio, llama.cpp, GGUF, Ollama
Internal link guidance
- Want to see local AI model running? Watch: How to run local AI models on M4 Mac with LM Studio: A complete 30-minute tutorial
- See who is making money using AI tools: He used Claude Code + AWS to build AI SaaS, and his monthly income was $12,000 for 3 months
- Open source alternative: Claude Code + DeepSeek V4 Building Tutorial: API Fees Dropped by 90% (15 Minutes)
Monetization angle
How can you make money from this trend?
WayToClawEarn focuses on verified earn playbooks—not just news. Start from these cases.
n8n + OpenAI affiliate site
Automate content and affiliate monetization
Claude + n8n automation agency
Charge monthly for agent workflow builds