Artificial intelligence development is entering a new phase in which AI models are becoming more capable, autonomous and difficult to evaluate.
That has prompted OpenAI chief scientist Jakub Pachocki to call for greater caution around the pace of AI development.
Pachocki has argued that AI developers should consider voluntary slowdowns while researchers work toward shared safety standards and better methods for monitoring increasingly capable systems.
The discussion is particularly relevant to cryptocurrency because AI agents are increasingly being considered for automated trading, blockchain transactions, wallet management and decentralized applications.
Why AI Safety Is Becoming More Difficult
Earlier AI systems were relatively easy to evaluate.
A chatbot generated a response, and researchers could inspect the output.
More advanced systems are different.
Modern AI agents can perform multiple actions, interact with software and pursue objectives over longer periods.
That creates a monitoring problem.
Instead of evaluating one response, developers may need to understand a chain of decisions.
An agent might:
- interpret a task,
- make a plan,
- access tools,
- execute actions,
- evaluate the result,
- change its strategy.
Each stage introduces new possibilities for failure.
Monitoring Model Reasoning
A major challenge is understanding how advanced AI systems reach their conclusions.
Researchers increasingly want to know not merely whether an AI produces a correct answer, but whether it follows safe and controllable processes to reach that answer.
That becomes more difficult as models become more sophisticated.
Pachocki has argued that monitoring increasingly capable systems can become less reliable without better shared standards.
This is important because many applications rely on AI systems behaving predictably.
Why This Matters to Crypto
The cryptocurrency industry is experimenting with AI agents.
Potential applications include:
- portfolio management
- trading
- automated DeFi strategies
- transaction execution
- blockchain analytics
- autonomous software agents
- wallet management
An AI agent connected to a blockchain could potentially execute transactions without asking a human for permission every time.
That creates efficiency.
But it also creates risk.
If the agent is compromised, manipulated or simply makes an incorrect decision, the resulting transaction may be irreversible.
AI Agents and Irreversible Transactions
Traditional software can often be corrected.
A mistaken email can be deleted.
A banking transaction may sometimes be reversed.
Blockchain transactions are different.
Once a cryptocurrency transfer is confirmed, recovery may be impossible.
That makes AI-powered blockchain systems particularly sensitive.
Developers therefore need stronger security controls around:
- transaction limits
- permissions
- identity
- smart-contract interactions
- wallet access
- automated decision-making
The Speed Problem
Another issue is development speed.
AI companies are competing aggressively.
Each company has incentives to release more powerful systems quickly.
That competition can create a difficult balance between innovation and safety.
If one company slows down while competitors continue advancing, it may fear losing market share.
This is why voluntary coordination becomes difficult.
Shared safety standards could potentially reduce that competitive pressure.
What Could a Slowdown Look Like?
A slowdown does not necessarily mean stopping AI development.
It could mean:
- more extensive safety testing
- additional model evaluations
- stronger deployment controls
- slower rollout of autonomous features
- improved monitoring
- more transparency
- coordination between developers
The goal would be to improve safety without freezing innovation.
AI and Blockchain Will Increasingly Intersect
The connection between AI and crypto is likely to expand.
Blockchain systems can provide digital ownership, programmable payments and decentralized infrastructure.
AI agents can provide automation and decision-making.
Combining the two could produce autonomous systems capable of:
- purchasing services
- managing digital assets
- interacting with decentralized applications
- paying other agents
- completing multi-step financial tasks
But this also means mistakes can have direct financial consequences.
The Next Security Layer
Crypto security has historically focused on:
- private keys
- smart contracts
- exchanges
- bridges
- phishing
- custody
AI agents introduce another layer:
decision security.
Even if the wallet and smart contract are secure, an AI agent could make a bad decision.
That means future blockchain infrastructure may require systems that can verify not just transactions but also the conditions under which automated agents are allowed to execute them.
Will AI Development Slow Down?
There is no simple answer.
The competitive incentives remain extremely strong.
Companies want better models.
Investors want growth.
Users want more capable tools.
Governments want technological leadership.
At the same time, increasingly autonomous systems create new risks.
The debate is therefore likely to continue between two competing priorities:
innovation speed
and
safety assurance
Final Takeaway
OpenAI chief scientist Jakub Pachocki’s call for greater caution highlights a major challenge facing the AI industry: as systems become more autonomous, evaluating whether they are behaving safely becomes more complicated.
That issue is especially important for crypto.
Blockchain systems provide powerful automation, but financial transactions are often difficult or impossible to reverse.
As AI agents gain the ability to control digital assets, AI safety will increasingly become a cryptocurrency security issue.
The next generation of crypto infrastructure may therefore need to solve two problems simultaneously:
Can AI act?
and
Can we safely control what it does?