For years, enterprise AI risk centered on the model itself: accuracy, hallucinations, bias, privacy, and security. That is changing.
Two developments this week show why.
Major banks are restricting third-party AI agents because existing identity controls cannot always distinguish human intent from autonomous software using valid credentials.
At the same time, pressure across specialized AI cloud providers is raising another concern:
What happens when the infrastructure supporting your AI workloads changes, consolidates, or fails?
The risks are different, but the lesson is the same:
AI adoption is moving faster than many organizations can govern it.
1. AI agents are challenging identity controls
Banks including JPMorgan Chase, Bank of America, Wells Fargo, Citi, and Morgan Stanley have restricted some third-party AI-agent activity.
In many cases, customers voluntarily granted access. The challenge begins after authentication.
Was the next action taken by the customer?
By an agent?
And was that action still within the customer’s intended scope?
Morgan Stanley has said it has not seen rogue agents accessing its platforms, an important distinction from more alarming headlines.
The real issue is more fundamental:
Valid credentials no longer guarantee that the person who authenticated is the one taking the action.
Wells Fargo’s upcoming Online Access Agreement reflects that shift. Beginning November 24, actions initiated by an authorized third-party AI agent may be treated as actions authorized by the customer, subject to applicable law.
That changes the identity equation.
Traditional model:
Authenticated user = authorized activity
Agentic model:
Authenticated user + verified agent + defined scope + runtime controls = authorized activity
Security teams now need to know:
which agent is acting
who authorized it
what it can access
what actions it can take
whether it exceeded its scope
how quickly access can be revoked
Mastercard’s Agent Pay framework points toward this emerging model, using registered agents, tokenized credentials, permissioning, and verifiable user intent.
2. AI infrastructure is becoming a resilience issue
The second risk sits lower in the stack.
Elon Musk warned this week that more than 200 specialized AI cloud providers, often called neoclouds, could face consolidation or failure as financing becomes more difficult.
That number is Musk’s estimate, not an independently verified forecast.
But the underlying pressure is real.
Gartner predicts that more than 80% of neocloud providers could cease operating independently by 2030 because of cost pressure and limited differentiation.
That does not mean all will disappear. Some may be acquired, merged, or absorbed into larger platforms.
Gartner also estimates that more than $3.5 trillion in AI infrastructure spending could be exposed to impairment by 2030 if current investment patterns continue.
For enterprise leaders, the question is no longer simply:
Can we run the workload?
It is also:
Can we move it if the provider changes?
The common thread
The banking story is about control risk. The neocloud story is about dependency risk. Both show how an AI initiative can fail even when the model works. That can happen when:
agents have too much authority
providers become unstable
workloads cannot move easily
security teams lack visibility
data becomes tightly coupled to one environment
infrastructure spending outpaces business value
Four questions leaders should ask now
1. Can we identify every non-human actor?
Agents should have their own identities, permissions, and audit trails.
2. What happens if a provider changes or disappears?
Review vendor viability, exit clauses, data portability, and migration plans.
3. How portable are our AI workloads?
Know what can move across hyperscalers, neoclouds, private environments, and edge infrastructure without major redesign.
4. Can we tie AI spending to measurable value?
Every major AI initiative should have a clear business outcome, owner, and success measure. More compute does not automatically create more value.
The takeaway
AI risk is moving beyond the model. It now includes:
Who is acting.
What they can do.
Where the workload runs.
How dependent you are on the provider.
Whether the economics still make sense.
The organizations best positioned for the next phase of AI will not necessarily deploy the most agents or buy the most compute. They will be the ones that can govern autonomy, manage dependencies, and adapt when conditions change.
Is your organization ready to scale AI?
CloudBait Navigator helps organizations evaluate readiness across AI, cloud, data, governance, security, and execution before technology decisions become expensive dependencies.
The question is not simply whether you can deploy AI. It is whether you are ready to operate it at scale.