Recently, IBM and NVIDIA announced an expanded collaboration focused on securing the next generation of AI agents. At first glance, this may look like just another AI partnership announcement. However, I believe the most important question behind this collaboration is not about AI performance.
It is about authority.
As generative AI evolved, most systems were designed to answer questions, summarize information, and assist users. AI agents are different. They are designed to take actions.
An AI agent may soon be capable of:
- Sending emails
- Creating purchase orders
- Accessing ERP systems
- Analyzing manufacturing data
- Modifying configuration files
- Executing business workflows
At that point, the key question changes.
It is no longer:
“What can AI do?”
Instead, it becomes:
“What should AI be allowed to do?”
We Do Not Give Every Employee Unlimited Access
Consider how enterprises operate today.
When a new employee joins a company, they are not granted access to every database, financial record, customer account, and production system.
Access is granted based on need.
The same principle should apply to AI agents.
Imagine an AI agent deployed inside a semiconductor company.
It may be allowed to:
✅ Read yield reports
✅ Analyze SPC data
✅ Generate defect classifications
✅ Recommend process improvements
But should it be allowed to:
❌ Change process recipes?
❌ Release production lots?
❌ Approve design modifications?
That is an entirely different discussion.
The Biggest Risk Is Not Intelligence
Many people focus on AI hallucinations.
That concern is valid.
However, in enterprise environments, excessive authority may become an even greater risk.
For example:
A user asks an AI agent to analyze inventory.
Instead of producing a report, the agent mistakenly places a purchase order.
Or consider an engineering review.
Instead of suggesting modifications, the AI directly changes design files.
The problem is not the intelligence of the AI.
The problem is authorization.
Why IBM and NVIDIA Are Focusing on Security
This is where the IBM-NVIDIA collaboration becomes interesting.
The underlying idea is simple:
AI agents should have identities, permissions, and accountability just like human employees.
Organizations must answer basic questions:
- Who is this AI agent?
- What systems can it access?
- What actions can it perform?
- Who monitors its decisions?
- How are its activities audited?
In other words, the industry is extending the concept of Zero Trust security into the age of AI agents.
Why This Matters for Semiconductor Manufacturing
The semiconductor industry handles some of the world’s most valuable information:
- Design IP
- Process recipes
- Reliability data
- Yield analysis
- Customer roadmaps
AI agents will increasingly participate in:
- Yield optimization
- Defect detection
- SPC monitoring
- Process control
- Design rule verification
But there is a significant difference between providing recommendations and making decisions.
An AI agent may be excellent at identifying a yield issue. That does not necessarily mean it should be allowed to modify a production process without human approval.
The smarter AI becomes, the more important governance becomes.
Conclusion
The most important question in the AI era may not be:
“Which AI model is the smartest?”
Instead, it may be:
“How much authority should we give AI?”
The IBM-NVIDIA collaboration highlights a reality that many organizations are only beginning to recognize.
The future challenge is not merely building more capable AI agents.
It is building AI agents that can be trusted.
In the coming decade, the companies that succeed may not be those with the most intelligent AI systems, but those with the most secure, accountable, and well-governed AI systems.
Ultimately, AI security is no longer just about protecting data.
It is about managing the digital employees that will increasingly work alongside us.