Pegasus4i Library · Essay

How an agent earns its way into IBM i

Calling a model from RPG is the easy part. The hard part is deciding what an agent is allowed to do, and proving who answers for it when it does. Here is how we answer that on a live IBM i, one rung at a time.

I

Projects fail at the authority.

I spent a decade as CTO of a regulated IBM i company. In that seat, nobody ever asked me whether a new system could connect to our core. Of course it could. The question was always the same: what will it be allowed to do, and who signs for it when something goes wrong?

AI agents bring that question back, and they bring it at a scale we haven't seen before.

The connection itself is no longer interesting. Db2 for i speaks SQL and JSON. IBM i can make HTTP calls from SQL. A gateway on a Linux partition can call RPG in a few lines. Any competent IBM i team can wire a model to the system of record in an afternoon. That isn't where projects fail.

They fail at the authority. Across the industry, the numbers say we are handing agents more than we can see:

80%
of companies say their AI agents have already taken unintended actions. The most common: reaching systems they weren't authorized to use.
SailPoint · 353 respondents · 2025
85%
of large companies using agentic AI have agents that act without a person involved in real time.
EY · 202 senior AI executives · Sep 2026
12.5%
can consistently trace an agent's complete activity. Only 17.5% can automatically stop one that starts behaving unexpectedly.
Salt Security · 2H 2026
97%
of organizations that reported a breach of an AI model or application had no AI access controls in place.
IBM · Cost of a Data Breach 2025

And the authority is moving faster than the oversight. In Illumio's survey of 1,100 IT and security decision-makers in fall 2026, 70% said they now let AI make decisions that needed a person's approval a year ago. Only 49% can fully see and audit every action an agent takes, and 24% said AI had reached data it shouldn't have.

Our own community is not ahead of this. In the COMMON × Fortra survey of the IBM i community (147 respondents, final results October 2026), 93% of respondents use at least one AI tool, and 15.6% have already built or deployed agentic systems. 44% have no security or governance framework, and only 7% have a comprehensive, active one. Asked for their single largest barrier to AI, more chose security or compliance concerns than anything else.

How prepared IBM i shops say they are to govern AICOMMON × Fortra survey, 147 respondents. Not prepared, no framework: 19.7%. Aware it's a concern, no plan: 23.8%. Some informal protocols: 26.5%. Documented governance protocols: 23.1%. Comprehensive framework, established and active: 6.8%. SECURITY AND COMPLIANCE GOVERNANCE FOR AI Not prepared, no framework19.7% Aware it's a concern, no plan23.8% Some informal protocols26.5% Documented protocols23.1% Comprehensive and active6.8% 44%
Fig. 1 · How prepared IBM i shops say they are to govern AI
COMMON × Fortra, The State of AI in the IBM i Community, 147 respondents, final results October 2026. Shares of all respondents.

So here is the way I think about it, and the way we build. An agent doesn't get access to IBM i. It earns it, one class of action at a time. Autonomy starts where mistakes are cheap and works its way toward the actions that matter. Every step up brings its own controls.

We run this on Silver Lake Financial Group, a fictitious lender we built so the ideas would have to survive contact with a real system. The IBM i under it is real. Below, the controls come first, as the rules we hold ourselves to; what's under "On Silver Lake" runs today. I showed an earlier version of this crossing live in my session at IBM TechXchange 2026 in Atlanta; the session deck is online.

Silver Lake Financial Group's home page: 'Thirty years of lending precision. A prequal answer in seconds,' with a check-your-rate panel of sample applicants. A banner across the top says Silver Lake is a fictitious company and accepts no applications.
Silver Lake Financial GroupA fictitious lender we built. Its site is live; its traffic is test traffic.
LoanIQSilver Lake's lending rules: RPG programs and Db2 for i, on IBM i on IBM Power Virtual Server. It makes every decision.
The LoanIQ gatewayThe only door to LoanIQ, on a Linux virtual server in IBM Cloud, reaching IBM i over a private connection. It runs as its own IBM i profile.
FootholdSilver Lake's credit agent. It calls no model, and sends each application through the gateway.
The credit deskNamed people who decide referrals, each under their own sign-in.
Where to looksilverlakefinancialgroup.com
Meet Silver Lake · The cast you'll meet in every "On Silver Lake" box below.
II

The ladder.

The ladder of agent authorityFour rungs, rising with what is at stake. Rung 1, read and inform: the agent, logged. Rung 2, check and prequalify: the agent, logged. Rung 3, decide: a deterministic program, no model. Rung 4, bind to terms and move money: a named person approves. Above them, overriding a governed decision has no rung and no door. NO RUNG · NO DOOR Override a governed decision RUNG 1Read and informAgent, logged RUNG 2Check andprequalifyAgent, logged RUNG 3DecideA deterministicprogram. No modelmakes the call RUNG 4Bind to terms,move moneyA named personapproves CHEAP TO GET WRONGBINDS THE COMPANY
Fig. 2 · Classes of action, rising with what's at stake. Who decides is fixed by the rung, not by the model.
RungWhat the agent doesReversible?Binds anyone?Who decides
1Read and informNot applicableNoAgent, logged
2Check and prequalifyFullyNoAgent, logged
3Decide, such as underwriteBy compensating entryConditionallyA deterministic program. No model makes the call
4Bind to terms, move moneyHard or costlyYesA named person approves
—Override a governed decisionNot applicableNot applicableNobody. No agent gets a door for it

The rungs are classes of action, not autonomy levels. Separately, every agent earns its level (read only, recommend, act with sign-off, act within limits) on its own record. The rung says how much is at stake; the level says how much the agent is trusted with it.

For each rung, three questions: what can go wrong, what stops it, and what it looks like on Silver Lake.

Rung 1

Read and inform.

What can go wrong

The agent sees more than it needs, and what it sees leaves the building. This is the most common failure in the field, not the most exotic one. In a 1Password survey of 1,000 security and engineering staff in mid-2026, 71% said their agents can reach sensitive information, and about four in ten organizations said their agents reach data outside what anyone approved.

OWASP, the nonprofit security community whose Top 10 lists are the industry's standard checklists, has a name for this: excessive agency. Its three root causes are excessive functionality, excessive permissions and excessive autonomy. Notice that none of the three is about the model. They are all about what we wrapped around it.

Excessive functionalityThe agent can reach more than its job needs.On IBM i: procedures, files and commands beyond its one task.
Excessive permissionsWhat it can reach, it can do too much with.On IBM i: a profile with special authorities or broad object authority.
Excessive autonomyIt acts with no one checking.On IBM i: it writes and commits with no rule or person in between.
Not on the list: the model.
Explainer · OWASP's three root causes of excessive agency (LLM06:2025), and what each looks like on IBM i.

What stops it

On Silver Lake ↑

The LoanIQ gateway runs under its own IBM i profile, with no special authorities. When it tried to read the decision table directly, IBM i refused, and the refusal was logged. The only door it has is the procedure it was granted.

Rung 2

Check and prequalify.

What can go wrong

A confident, wrong answer flows straight into a process. Or a document, an email or a web page tells the agent to do something else, and it does. In that same 1Password survey, 47% of developers said an agent had acted unintentionally after following instructions hidden in a web page, document, email or tool output.

The UK's National Cyber Security Centre has been blunt about this. Prompt injection may never be fully mitigated the way SQL injection was, because a language model draws no real line between instructions and data. Its advice is to reduce the impact. OWASP agrees on the order of things: its first Top 10 for agentic applications, published in December 2025, puts agent goal hijack at number one.

Stop asking how to make the model safe. Ask what the model's answer is allowed to touch.

What stops it

An answer with a shape, checked twiceThe model proposes. It returns a typed answer: a decision from a closed list, a reason code and a review flag. The gateway checks it against the business rules, then the IBM i program checks it again before acting. Anything that fails either check gets the safe default, usually a person, and the refusal is recorded with its reason. MODELProposes TYPED ANSWERdecision · reason code· review flag GATEWAYchecks businessrules IBM ichecks again,final say ACTon therecord FAILSFAILS Safe default, usually a personrefusal recorded with its reason THE MODEL NEVER HANDS FREE TEXT TO A PROGRAM THAT ACTS
Fig. 3 · Answers have a shape. The gateway checks, then the system of record checks again.
On Silver Lake ↑

Silver Lake's credit agent, Foothold, calls no model, so it shows the other two controls. Every call to the LoanIQ gateway carries a service token for its caller. A caller without one is refused at the network edge, the gateway checks the identity again, and the caller's name is recorded on the IBM i job and on the run. Foothold passed its first gate review on the evidence: its outbound connections are limited to a fixed list enforced in code, a test proves an unlisted destination is refused and logged, and no applicant data appears in its traces or logs. And when we revoked Foothold's credential at the edge, its next call was refused before it reached the gateway. The applicant was told it isn't a decline, never given a default approval.

Rung 3

Decide.

What can go wrong

A model makes a regulated decision. Nobody can explain the reasons to an examiner, and nobody can reproduce it next month.

What stops it

The model never decides. A deterministic program decides and records its reasons. Where a decision is regulated, this isn't a preference. It's the only design that survives an exam.

Corrections are new entries, never edits. If a decision was wrong, you write a compensating entry and the history stays intact. And the tables behind the decision are owned by a profile no agent or service uses, so no agent can change the rules that guard them.

On Silver Lake ↑

Foothold sends each application through the LoanIQ gateway, which calls RPG on IBM i. The program decides, and the decision is recorded in Db2 for i, written only through the gateway's one procedure. The first real decision, on test traffic, went through on November 10.

Rung 4

Bind and move money.

What can go wrong

An agent commits the company to terms, or moves money, on authority nobody actually gave it. Or the authority was real but stale: approved yesterday for a different amount. And governance bends under a deadline: in EY's 2026 survey, 47% of senior AI executives said their organization had skipped its own AI governance process for an urgent deployment.

What stops it

An approval is a key that fits one lock, onceApproved by a named person, never the requester. Bound to the exact customer and the exact amount. Expires after a window the owner sets. Used once, in the same transaction as the change and its record. If any part fails, everything rolls back and the approval is still there. 01020304 ApprovedBoundExpiresUsed once by a named person,never the requester to the exact customerand the exact amount after a windowthe owner sets in the same transactionas the change and record If any part fails, all of it rolls back, and the approval is still there to try again.
Fig. 4 · One key, one lock, once. The rules for rung 4; not every one runs on Silver Lake yet.
Silver Lake Financial Group's marketplace page: 'Send your agent shopping. It can fill the cart. It can't sign for you.' A banner across the top says Silver Lake and its partner merchant are fictitious and the store runs in test mode.
Fig. 5 · Silver Lake's marketplace: financing at checkout, decided on IBM i. The agent can fill the cart; it can't sign. Silver Lake and its partner store are fictitious, and the store runs in test mode.
On Silver Lake ↑

On November 11 we ran this rung end to end for the first time. A fictitious shopper asked for $8,280. LoanIQ on IBM i referred it to Silver Lake's credit desk, where a named person approved it under their own sign-in; the gateway takes that name only from the verified sign-in, ignoring anything in the request. Foothold then showed one offer at LoanIQ's own terms, and the shopper accepted it.

An offer can be accepted only at the terms the lender quoted, once per decision: a rate or term that doesn't match is refused, a second acceptance returns the first, and a decision that was declined or is still referred can't be accepted. The IBM i holds the referral, the desk decision and the acceptance under one decision number, and a gateway restart loses none of them. No money moves. Not every rule above is shown running yet, and I'd rather say so: the approval's expiry window, for one.

—

The rung no agent climbs.

There is one action on the ladder with no rung under it: overriding a governed decision. No agent gets a door for it. No procedure exists for it to call.

You can't misuse a door that was never built.
III

What's different on IBM i.

People ask me whether any of this is IBM i–specific. Most of the ideas aren't. The ladder works on any platform. What's different is where the controls live, and how much you have to build yourself.

On a typical cloud or Linux stack, an agent's identity lives in an identity provider, its permissions in an access policy, its audit trail in a separate logging service, and its transactions spread across services that each own their own database. Each piece is good. They just aren't the same piece. When a business action spans services, there's no single commit. You design compensating steps by hand, and cloud architecture guidance itself warns that those compensations might not always succeed.

On IBM i, the controls are part of the operating system and the database:

Where the controls liveOn a typical cloud or Linux stack, identity, permissions, the narrow door, the audit trail and each service's database are separate pieces joined by seams you build and watch. On IBM i, the same controls sit inside one system: the operating system holds the user profile, adopted authority and the audit journal; Db2 for i holds commitment control, constraints, triggers and grants, and row and column access control. TYPICAL CLOUD / LINUX STACK IBM i IDENTITYidentity providerPERMISSIONSaccess policy engineNARROW DOORAPI gateway + app codeAUDIT TRAILseparate log pipelineSERVICE Aits own databaseSERVICE Bits own database OPERATING SYSTEMUser profile, checked on every objectAdopted authority, the narrow doorQAUDJRN, who did whatDb2 for iCommitment control, one unit of workConstraints, triggers and grantsRCAC, row and column limits EVERY RED SEAM IS ONE YOU BUILD AND WATCHONE SYSTEM · ONE COMMIT
Fig. 6 · Where the controls live. On a typical stack you assemble them; on IBM i they come with the operating system and the database.
What an agent needsTypical cloud / Linux stackIBM i
Its own identityA service account or workload identity in an identity provider, separate from the dataA user profile, checked by the OS on every object it touches
A narrow doorAn API gateway plus app code that enforces scopeA procedure that adopts its owner's authority; the agent can call it but never touch what's behind it
All-or-nothing changesHard across services; sagas with hand-built compensationsCommitment control across Db2 for i in one unit of work
Rules that can't be bypassedApplication code, policy enginesConstraints, triggers and grants owned by a profile no agent uses
A record of who did whatSeparate log pipeline, often per serviceThe security audit journal (QAUDJRN), built in; plus journaling on every table that matters
Row- and column-level limitsRow-level security in the database (where supported)Row and column access control (RCAC) in Db2 for i

That is a real head start. The building code is already in the building.

But I want to be honest about two things, because a CTO who oversells this gets caught.

First, having the controls isn't the same as using them. Fortra's 2026 State of IBM i Security study (June 2026), covering 163 partitions, shows real progress: 96% of systems had the security audit journal on, and the share of users holding all-object authority fell to 8%. The same study found an average of 889 users per system holding job control authority, 10% of systems with default passwords on at least one profile, and only 31% with exit programs covering all 27 common network entry points. Every one of those is a path an agent could use. Agent governance on IBM i starts with the basics most shops already know they should finish.

Second, adopted authority cuts both ways. It's the strongest door IBM i offers, and it travels. IBM's own documentation explains that adopted authority can flow to programs further down the call stack. Design the owning profile and the call chain deliberately, or the narrow door turns into a wide hallway.

IV

The part most guidance stops before.

Everything above is about the moment of crossing. The harder problem starts after launch.

Agents don't run on IBM i. They run on Linux beside it, or in a cloud, with their own performance, security and cost behavior. So a shop that goes agentic now runs two systems that know nothing about each other: the IBM i it has trusted for decades, and an agent layer it has never had to oversee.

The scale is coming fast. Palo Alto Networks' 2026 Identity Security Landscape puts machine identities at 109 for every human. And in Salt Security's 2H 2026 survey, more than 60% said reconstructing an agent's activity from the day before would take hours or days of manual work. In EY's survey, 26% of executives at companies using agentic AI said they can't detect an unauthorized agent running inside their own organization.

On Silver Lake, one correlation ID sits on the agent's run, on the gateway's run and on the Db2 for i decision record, and the gateway's trace ID is on the IBM i job that did the work. One request reconstructs as one trail. And none of our agents acts on its own without a stop we have tested and recorded.

One request, one trail, across two systemsOn the agent side, on Linux: the agent's run and the gateway's run. On IBM i: the job that did the work and the Db2 for i decision record. One correlation ID sits on the agent's run, the gateway's run and the decision record, and the gateway's trace ID is on the IBM i job. AGENT SIDE · LINUX IBM i AGENTAgent run GATEWAYGateway run IBM i JOBDid the work Db2 for iDecision record ONE CORRELATION ID trace ID on the job
Fig. 7 · On Silver Lake, one request reconstructs as one trail across both systems. Test traffic on a fictitious lender.

Behind that sits a simple principle: you should be able to see both systems, the platform and the agents, in one view. Try these on your own shop:

We built the view that answers those. It's called OneLens4i. It runs today on our own agents and our own IBM i; no client runs it yet.

V

Built in, not bolted on.

Everything on the ladder uses what an IBM i shop already owns. None of it requires a new platform. All of it requires deciding, in advance, what each agent is allowed to touch.

Bolted on versus built inBolted on: an agent layer is strapped to the outside of the watch, with its own identity, rules and log, joined to IBM i by a shared login, keeping its own time. Built in: the case is open and the agent is a small gear meshed with the IBM i movement through one door, its own user profile and one procedure, so commitment control and the audit journal keep one time for both. BOLTED ON IBM i AGENT LAYER own identityown rulesown logown clock SHARED LOGIN strapped to the case, keeping its own time BUILT IN IBM i AGENT one door: its own profile, one procedure commitment control,the audit journal one movement, keeping one time
Fig. 8 · Bolted on, the agent brings its own rules. Built in, it lives by the platform's. More on why in You can't bolt onto a watch (watch the video) and The Back of the Watch.

You can start that this week, before any agent goes near production:

  1. List every agent and AI tool that touches your IBM i, and the profile each one signs on as. If two share a profile, or one signs on as a person, that's your first fix.
  2. Put each one on a rung, and give it nothing above that rung: no procedure, no grant.
  3. Finish the basics: default passwords, *PUBLIC access, open servers and exit points. Fortra's numbers in section III say most shops still have at least one of these gaps.

Not sure where your shop stands? The Agent-Ready Read asks eight questions at the level you run the box, takes about three minutes, and needs no account.

The models will change every few months; I track how fast in The Premium Has a Half-Life. The rules you build into your own platform stay yours.

That has always been the IBM i bargain. AI didn't change it. It just raised the stakes.

Silver Lake Financial Group is a fictitious company created by Pegasus4i for demonstration; its traffic is test traffic. Pegasus4i is an IBM Business Partner.

Own your source. Own your intelligence.
Sources
  1. SailPoint, AI Agents: The New Attack Surface (Dimensional Research, 353 respondents), May 28, 2025. SailPoint press release
  2. Salt Security, State of Agentic AI and API Security 2H 2026 (380+ security and technology professionals). Campus Technology, Oct 21, 2026
  3. IBM, Cost of a Data Breach Report 2025. IBM Newsroom, Jul 30, 2025
  4. COMMON × Fortra, The State of AI in the IBM i Community (147 respondents, final results October 2026). signal4i.ai/common-ai-results
  5. 1Password, AI agent governance survey (1,000 security and engineering staff at large U.S. companies, late May–early June 2026). Help Net Security, Jul 29, 2026
  6. OWASP, LLM06:2025 Excessive Agency
  7. UK National Cyber Security Centre on prompt injection (Dec 2025). Coverage: Security Boulevard; Malwarebytes
  8. Saga distributed transactions pattern, Azure Architecture Center
  9. Fortra, 2026 State of IBM i Security Study (163 partitions). IT Jungle, Jun 8, 2026
  10. IBM, Objects that adopt the owner's authority, IBM i documentation; Adopted authority, IBM Support
  11. EY, survey of 202 senior AI executives at public companies with at least $1 billion in revenue, fielded May 28–June 15, 2026. EY newsroom, Sep 2026
  12. Illumio, survey of 1,100 IT and security decision-makers in 10 markets, 2026. IT Brief, Oct 2026
  13. OWASP GenAI Security Project, Top 10 for Agentic Applications for 2026, Dec 9, 2025
  14. Palo Alto Networks, 2026 Identity Security Landscape. Help Net Security, May 14, 2026
Essay · Pegasus4i Library · ← Speed to Intelligence ↑ The 4i Library