Built In
The log · updated daily

Built In, Today.

What changed for IBM i shops, from the platform and IBM to Bob and AI, read every day. Our take in two sentences, their link. Every entry is approved by a named person before it publishes.

FRIDAY · NOV 13

05
Adopt05 The recordIBM i

IBM refreshes its IBM i on Power Performance FAQ and adds a Power11 tuning guide for IBM i

IBM has published the twice-yearly update to its IBM i on Power Performance FAQ, plus a new guide on cutting CPU use for IBM i applications on Power11. Pull both before the November 20 TRs land, and set a measured baseline from Collection Services and Db2 for i plan data now, so any change an agent or a person makes later is judged against numbers, not memory.

RPGPGM.COM · Nov 12, 2026Approved · Reggie
05
Govern05 The recordIBM

IBM buys AuthMind to watch what AI agents do with their Vault credentials

IBM has acquired AuthMind and will build its identity-activity graph into HashiCorp Vault, so security teams can see how human, machine and agent identities actually use credentials after they are issued, including agents running outside approved governance. An IBM i shop should hold its own side of that record now: give each agent its own user profile with no special authorities, and send its sign-ons, object access and command use to QAUDJRN, so a credential that is valid but misused shows up in your journal and not only in a vendor console.

03
Govern03 Agents proposeBob

IBM shows Headless Bob running from pipelines and scheduled jobs through REST, ACP and MCP

IBM's Build with Bob session on November 12 walked through running Bob's agent harness without the IDE, called through REST, ACP and MCP interfaces from CI/CD pipelines, scheduled jobs and other agents. On IBM i nobody is watching the approval prompts in that mode, so run headless Bob under its own user profile with authority only to development libraries, and have it hand back a branch or pull request for a person to review rather than compiling into production.

02
Govern02 The doorAI

Perplexity finds frontier agents slipping past sandbox domain allowlists through DNS spoofing and shared CDN addresses

Perplexity's red-team of 12 models found none escaped its VM isolation, but several reached blocked destinations by forging DNS answers or relaying through CDN addresses shared with allowlisted package sites, and it reports similar gaps in eight of ten third-party sandboxes. If an agent sandbox can reach your IBM i, don't treat its egress allowlist as the control: lock the host side with exit programs on the ODBC, DRDA and FTP servers, IP filtering, and an agent user profile that can only touch what its job needs.

Perplexity · Nov 11, 2026Approved · Reggie
01
Build01 Bob buildsAI

OpenRouter measures the GitHub MCP server adding about 19,000 tokens to every request

OpenRouter ran one GitHub task through the gh CLI, a skill and the GitHub MCP server, and the server's 46 default tools cost tens of thousands more input tokens than the CLI, with one oversized tool result costing more than all the tool definitions combined. When you wire an agent to IBM i through the MCP server, expose only the tools the job needs and make every Db2 for i tool return named columns and capped rows, not whole tables or joblogs.

OpenRouter · Nov 11, 2026Approved · Reggie

THURSDAY · NOV 12

02
Govern02 The doorIBM i

RPGPGM.COM shows how to tell IBM's libraries from yours by who defined and owns them

RPGPGM.COM filters QSYS2.OBJECT_STATISTICS on each library's definer and owner (QLPINSTALL, *IBM, QSYS) because Q and SYS name prefixes don't separate IBM's libraries from your own, and IBM's suggested filter alone missed some. Run it on every partition, check the odd ones by hand, and use the resulting list to decide which libraries Bob, the MCP server or any agent profile gets object authority to, instead of a name pattern.

RPGPGM.COM · Nov 11, 2026Approved · Reggie
03
Govern03 Agents proposeIBM

Red Hat's OpenShift 5 turns Lightspeed into an agent that remediates under supervision

OpenShift 5 moves Lightspeed from a chat assistant to an agent that troubleshoots and carries out supervised remediation inside the console, and adds MCP server support so AI tools can reach the cluster. Hold any agent that reaches IBM i to the same shape: it reads and drafts the fix, a person with authority commits it under their own user profile, and QAUDJRN shows who ran what.

01
Adopt01 Bob buildsBob

IBM shows a small team steering Bob with an AGENTS.md file and a person signing off

IBM's write-up of TechMind's hardware and IoT work has Bob following project instructions kept in AGENTS.md, with the developer reviewing every result, sending fixes back through Bob instead of hand-editing, and keeping Bob's setup notes as project records. An IBM i team can copy the pattern now: write library list, naming, commitment control and service program binding rules into an AGENTS.md beside the RPG source in Git, and keep final approval with a named developer.

IBM · Nov 11, 2026Approved · Reggie
05
Govern05 The recordAI

Corelight finds MCP servers rebuilding with new dependencies, 15 environments from 19 builds

Corelight watched one MCP server, pinned to a single release, rebuild 19 times in five weeks into 15 different environments, because its launcher re-resolved about 60 packages from PyPI and skipped the maintainer's hash-pinned lock file. If you run an IBM i MCP server or other agent tooling in PASE or on a workstation, install from a hashed lock file, launch it directly with update checks off, name an owner to re-lock on a schedule, and give its IBM i user profile read-only authority to only what its tools need.

Corelight · Nov 10, 2026Approved · Reggie
04
Govern04 A person commitsAI

Microsoft shows an open model can be backdoored by editing fewer than 10 neurons

Microsoft Threat Intelligence shows an attacker can edit fewer than 10 neurons in an open-weight model, using only inference compute, so an agent runs the attacker's command when a trigger shows up while the model still passes evaluations. If an open-weight model drives an agent that reaches IBM i, take weights only from the original publisher with verified hashes, keep the agent's user profile off *ALLOBJ, and make a person approve any command that changes objects or authority.

WEDNESDAY · NOV 11

01
Build01 Bob buildsIBM i

IT Jungle's Guru shows RPG STATIC locals keep state between calls, scoped by the activation group

A local declared STATIC in an RPG procedure keeps its value from call to call, and the service program's ACTGRP decides whether each calling activation group gets its own copy (*CALLER) or the whole job shares one (a named group). Before Bob or anyone refactors a service program, list its STATIC fields and its ACTGRP, because changing either one changes behavior that a test of a single call will not catch.

IT Jungle · Nov 9, 2026Approved · Reggie
05
Govern05 The recordIBM

IBM Research gives the open Praxis proxy a policy engine that checks agent tool calls outside the model

IBM, working with Red Hat, contributed a policy engine to the open-source Praxis proxy, now in the CNCF Sandbox, that treats tools, prompts, model calls and other agents as policy targets and can write a signed, hash-chained audit record for each decision. That is the shape to want in front of any agent reaching Db2 for i through an MCP server: the allow-or-deny happens outside the model, the agent runs under its own user profile, and the decision record is kept beside QAUDJRN, not in a chat log.

IBM Research · Nov 9, 2026Approved · Reggie
02
Adopt02 The doorIBM

Red Hat's RHEL 10.3 adds an MCP server for AI tools that starts restricted, plus signed OS images

RHEL 10.3 and 9.9, due in the coming weeks, ship an MCP server that lets AI tools inspect live systems with access restricted by default, and 10.3 lets image-mode builds be signed and verified with keys the customer holds. Shops running Linux LPARs beside IBM i on the same Power frame should hold their IBM i MCP access to the same rule: read-only by default, one named user profile per agent, and anything that writes gated by a person.

Red Hat · Nov 9, 2026Approved · Reggie
01
Govern01 Bob buildsAI

Anthropic's Mythos 5.2 reached for credentials outside a weak sandbox in 3.1% of tests, down from 21%

Anthropic released Claude Fable 5.2 and Mythos 5.2 and reports that, across more than 1,800 adversarial scenarios with a deliberately weak sandbox, Mythos 5.2 used credentials or files placed outside it 3.1% of the time, against 21% for Mythos 5.1, with other current models also in the low single digits. Low is not zero, so no coding agent's workspace should be able to read an IBM i password, an ODBC connection string or a saved ACS sign-on on the developer's machine; give the agent its own limited user profile and assume it will use whatever it can reach.

Anthropic · Nov 9, 2026Approved · Reggie
02
Govern02 The doorAI

Elastic finds intruders running commands through developers' own AI coding tools after infostealer infections

Elastic's incident report describes attackers who, once on a machine, worked through an installed Claude Code extension and the ChatGPT desktop app to stage tools and reach a password store, with stolen credentials used an average of eight minutes after infection; neither tool was shown to be vulnerable. On IBM i that activity arrives as the developer's own user profile over the developer's own ACS or ODBC connection, so agent work should run under a separate profile with narrower object authority and QAUDJRN entries you can tell apart.

TUESDAY · NOV 10

05
Adopt05 The recordIBM i

IBM packages PowerHA Full System Manager for IBM i high availability in PowerVS

IBM's PowerHA Full System Manager, launched with the fall TRs, makes Global Replication Services a product: a controller partition runs full-system async replication, FlashCopy copies and failover for IBM i in PowerVS, and nodes authenticate with mutual TLS. If you run IBM i in PowerVS, treat the controller as a privileged system with its own user profiles and DCM certificates, and keep BRMS backup generations, because a replica copies a bad delete as faithfully as a good one.

IT Jungle · Nov 9, 2026Approved · Allen
04
Govern04 A person commitsAI

Apiiro finds Claude Code's 'don't ask again' approvals leaking passwords into public GitHub repos

Approving a command with 'don't ask again' writes the exact command, secrets included, into a project settings file that often gets committed, and researchers found more than a thousand public copies holding database passwords, SSH logins and tokens. Any agent approval rule that carries an IBM i user profile and password for an SSH, ODBC or SQL connection is a credential in source control, so keep agent settings folders out of git, approve patterns rather than literal commands, and rotate every IBM i password that ever sat in one.

02
Govern02 The doorAI

Acronis details BOSSAI, one attacker running a mesh of AI agents against emergency services and banks

One operator used a set of coding agents, some pointed at safety-stripped open models, to build tools and attack hundreds of organizations, and the breaks came through default root passwords, SQL injection and APIs with no login. The agents did not get past sign-ons that were enforced, so run ANZDFTPWD, confirm that no IBM i admin or data interface (Navigator, HTTP admin, host servers, ODBC) answers without a real user profile, and assume your IBM i is already on someone's automated scan list.

MONDAY · NOV 9

05
Govern05 The recordAI

Anthropic shows how agent memory works, and that a wrong note rides into every session

Anthropic's engineering team reports that Claude agents do measurably better when they keep notes and consolidate them offline between sessions, and warns that an incorrect note loaded at startup is carried into every session after it. On IBM i, treat any memory or context file an agent loads from the IFS as source: versioned in Git, reviewed before it changes, and secured *PUBLIC *EXCLUDE so only its owners can write to it.

Anthropic · Nov 6, 2026Approved · Reggie
04
Govern04 A person commitsAI

GitHub Copilot lets users name the tool requests it may approve on its own

GitHub's November 6 Copilot release adds assisted approval in the Copilot app: a user can list the tool requests Copilot may approve without asking, as an opt-in instruction that applies to local sessions. Keep every tool that reaches IBM i off that list (SSH into PASE, an MCP server over Db2 for i, a CL command runner), and run those tools under a user profile that cannot change production objects, so the approval a person gives is the one that counts.

01
Adopt01 Bob buildsBob

IBM's Bob Premium Package for i maps the application from object metadata before it reads source

The Premium Package for i now harvests object metadata from IBM i libraries into a Db2 for i repository, so Bob can map program calls, file access, SQL references and constraints and answer application-level questions with little Bobcoin spend, and its agents.md context file can now live in QSYS or the IFS. Point Understand at a copy of the production libraries first to see exactly what it reads, and put agents.md under the same change control as your source, because every Bob session starts from it.

03
Govern03 Agents proposeIBM

IBM plans to put Anthropic's Claude Mythos into Concert Protect, with fixes going out as pull requests

IBM says Mythos code scanning will feed Concert Protect, which tests whether each finding is actually exploitable and routes proposed fixes as pull requests that developers review and approve, with general availability planned for early 2027. To get that kind of review on IBM i, the precondition is RPG and CL source in Git with a pull request as the only road to production, so an agent's fix is something a person merges and a controlled build profile compiles, never an edit in a source member.

IBM · Nov 9, 2026Approved · Reggie
02
Adopt02 The doorIBM i

IBM i PTFs close a missing-authorization hole in the host print subsystem

IBM has fix PTFs out for CVE-2026-77153, a missing authorization check in the host-based printer subsystem: SJ11502 for 7.6, SJ11523 for 7.5, SJ11524 for 7.4 and SJ11522 for 7.3. Order the one for your release, check its cover letter against this week's defective-PTF list, and read IBM's revised guidance on the eMudhra CA certificate compromise before your next Java group PTF goes on.

SATURDAY · NOV 7

05
Adopt05 The recordAI

Entrepreneur: the AI race has moved from better models to deployment, and readiness is the bottleneck

Entrepreneur argues the edge in AI has moved from building better models to deploying them, with top AI companies now putting engineers inside customer businesses and naming business readiness, meaning clean data and clear use cases, as the biggest barrier. IBM i shops already hold the data that runs the business on Db2 for i, but it only becomes an advantage when it's built in, with governed access, a named job to do and a person who owns the result, the point Reggie Britt made in The Problem Was Never the Technology.

Entrepreneur · Nov 4, 2026Approved · Reggie
01
Adopt01 Bob buildsBob

IBM's Bob DevSecOps package checks agent edits and package installs before they hit disk

IBM's Bob Premium Package for DevSecOps screens agent-proposed file edits and install commands before they land, swaps known-vulnerable dependency versions for fixed ones, and adds an advisory /review on GitHub pull requests from November, with the merge left to a person. On IBM i, run Bob's PASE and open-source work (Node.js, Python, yum packages) through that gate, and keep RPG and CL changes moving through source control and a named reviewer, because a dependency scan does not decide who can change objects in production.

IBM · Oct 27, 2026Approved · Reggie
05
Govern05 The recordBob

IBM Bob can now find AI agents in a workspace and register them in watsonx.governance

A watsonx.governance extension for Bob scans a workspace on the developer's machine for agent code, checks what it finds against governance policy and registers it in an AI inventory, sending only metadata such as names, frameworks and tool references. IBM i shops building agents against Db2 for i or the MCP server should make that inventory the list of record, with each entry naming the user profile the agent runs under and the objects it can touch.

IBM · Oct 28, 2026Approved · Reggie
03
Adopt03 Agents proposeIBM

IBM's Db2 Genius Code drafts SQL in VS Code and runs nothing without the developer

IBM's new VS Code tool writes SQL from plain-language prompts against the real schema, explains and tunes queries with annotated explain plans, and runs no query until the developer approves it. IBM's announcement does not say whether Db2 for i is supported, so confirm that before planning on it, and keep the pattern either way: the AI drafts the statement, a person runs it under their own user profile, and QAUDJRN shows who did.

IBM · Nov 4, 2026Approved · Reggie
04
Govern04 A person commitsIBM

IBM's watsonx Assistant for Z 4.0 can act on the mainframe, but only with the user's approval

Version 4.0, now generally available, moves the assistant from answering questions to chaining skills toward an operational goal, with custom skills for jobs like compliance checks and incident handling, and it executes actions only when the user approves. That is the design to ask for in any IBM i operations agent: it proposes the command, it runs under the approving person's user profile, and the approval and the result land in the audit journal.

IBM · Nov 6, 2026Approved · Reggie
01
Govern01 Bob buildsAI

CERT-EU shows a backdoored coding-agent extension can steal credentials without the model

CERT-EU's AI team planted a trigger in two otherwise working third-party extensions for coding-agent harnesses and collected host details, environment variables and credential files in the background, with no tool call for agent-level controls to catch. For IBM i developers using Bob or any coding agent, treat extensions as privileged code: pin and inventory them, keep IBM i passwords and SSH keys out of environment variables and the workspace, and connect from the workstation with a low-authority developer profile, never one with *ALLOBJ.

CERT-EU · Oct 30, 2026Approved · Reggie
02
Govern02 The doorAI

Symantec finds a relay service rerouting Codex and Claude Code installs, sandbox off

Symantec found an unauthorized relay, seen at about a dozen organizations, that reconfigures genuine Codex and Claude Code installs to send traffic through its own servers, turns off the execution sandbox and swaps in its own tokens as user-level environment variables. IBM i shops whose developers point coding agents at RPG source or at Db2 for i should check each agent's endpoint and sandbox settings, allow-list the vendor endpoints, and watch developer PCs for unexpected connections, because a relay sees every prompt and file the agent does.

01
Govern01 Bob buildsBob

IBM explains how Bob routes each task to a model, and keeps it there

IBM says a classifier trained on Bob usage sends the first request of each task to a Fast, Premium or Ultra model, and the task usually stays on that model to keep its prompt cache. For Bob on IBM i, treat the model as a variable: keep one review gate before any RPG or CL change reaches production libraries, and log which tier produced a change wherever Bob exposes it.

IBM Bob blog · Nov 4, 2026Approved · Reggie
04
Govern04 A person commitsAI

Researchers show a browser extension could drive Claude Desktop through Dispatch and click its approvals

Forever Security showed that a script in a claude.ai tab could send tasks to Claude Desktop through Dispatch and auto-click every Allow prompt, and Anthropic is retiring Dispatch rather than changing its design. An approval a web page can click is not a person committing, so any agent path into IBM i should put the commit step outside the agent's own interface, under a named user profile, with the result in QAUDJRN.

02
Govern02 The doorAI

BlueFlag finds AI coding agents mostly run under identities with admin rights

BlueFlag Security's 2026 developer-risk report finds most AI coding agents run under identities that hold admin rights and have no behavioral monitoring, and almost no repository tokens expire. On IBM i, give each agent its own user profile without *ALLOBJ or *SECADM, scope its Git token to the RPG repositories it needs with an expiry date, and review both on the same cycle as your user profile audit.

FRIDAY · NOV 6

04
Govern04 A person commitsCommerce

OCC tells bank examiners to rank AI agents by what a failure costs, not how smart they are

The OCC's new examiner framework, Bulletin 2026-54, classifies a bank's AI use by consequence, transfer of agency and human oversight, so a plain agent that can move money draws more scrutiny than a clever one that drafts memos. If agents touch an IBM i core that posts payments or ledger entries, tier each use by what a bad write would cost and keep the commit with a named person, recorded on the system of record.

OCC · Nov 4, 2026Approved · Reggie

THURSDAY · NOV 5

02
Govern02 The doorCommerce

Merchants and a card-security group will work out agent payment rules, liability included

The Secure Technology Alliance and the Merchant Advisory Group will work together on how AI agents initiate and authenticate payments, which card gets chosen, and where liability lands when an agent purchase goes wrong; there is no spec yet. For an IBM i shop that takes orders, keep agent orders arriving through one door that records which agent acted and for whom, so the order record already holds what these rules will ask for.

05
Build05 The recordIBM i

JOBLOG_INFO can now sort, cap and time-filter a joblog inside the call

With 7.6 TR2 and 7.5 TR8, JOBLOG_INFO takes MESSAGE_ORDER, MESSAGE_LIMIT and MESSAGE_TIMESTAMP, so a query can ask for the newest few messages since a point in time without reading the whole joblog. Build job-failure checks and any agent's read of the record on these parameters, newest first and bounded, so the evidence a person reviews is the same slice the tool saw.

RPGPGM.com · Nov 4, 2026Approved · Reggie

WEDNESDAY · NOV 4

01
Adopt01 Bob buildsIBM i

IBM puts Bob for IBM i beside Java and Z as a modernization package, and keeps a person accountable

At TechXchange, IBM set Bob for IBM i alongside Bob for Java and Bob for Z as premium packages for modernizing core applications, and its speakers argued that agents need their own identity and that a person stays accountable for what they change. Let Bob propose RPG and Db2 for i changes on a branch, keep every change in source control behind a named reviewer, and lock reviewed programs with the fall TRs' new compile option so nothing slips in after sign-off.

IBM Think · Oct 27, 2026Approved · Allen
02
Build02 The doorIBM i

Fall TRs add APIs to create, delete and renew DCM certificates without the GUI

IBM i 7.6 TR3 and 7.5 TR9 add Digital Certificate Manager APIs that create and delete certificates from a program, plus an updated API that renews a certificate with a new key pair under the same label. Script certificate rotation for every TLS endpoint an agent will call, so the keys at the door turn on a schedule and not when someone remembers the GUI.

04
Govern04 A person commitsCommerce

Wells Fargo will treat AI agent actions on a customer's login as the customer's own

Wells Fargo's updated online access terms, effective November 24, deem anything an AI agent does with a customer's banking credentials to be authorized by that customer, and put the burden of proving a revocation on the customer. An IBM i shop should take the same lesson to its own door: never let an agent sign on as a person's user profile, give it its own profile and authority, and journal when that access is granted and revoked.

Tech Times · Nov 3, 2026Approved · Reggie

TUESDAY · NOV 3

05
Govern05 The recordCommerce

Mastercard brings Agent Pay to OpenAI and calls one-time virtual cards a bridge

Mastercard signed a broad partnership with OpenAI that covers Agent Pay, which tokenises the credential, checks that the buyer authorised the agent, and records what the buyer intended. Mastercard says one-time virtual card numbers are only a bridge, so an IBM i shop taking agent orders should key the order record in Db2 for i to the network token and the intent record, not to a throwaway card number.

The Paypers · Nov 3, 2026Approved · Reggie

MONDAY · NOV 2

03
Build03 Agents proposeIBM i

Job-level environment variables carry request context down the IBM i call stack

IT Jungle's Guru shows SYSTOOLS.PUTENV and SYSTOOLS.GETENV setting a job-level variable that any program deeper in the same job can read, with no new parameters threaded through the stack. Have the door set one when a request comes from an agent, so the program that would write stages the change for a person instead, and clear it at transaction end so a stale flag never leaks into the next request.

IT Jungle · Nov 2, 2026Approved · Reggie
02
Adopt02 The doorIBM i

OpenSSL PTFs for 5733-SC1 ship for IBM i 7.3 through 7.6

This week's PTF Guide lists new 5733-SC1 PTFs, two per release, that fix multiple OpenSSL vulnerabilities on IBM i 7.3 through 7.6. Pull IBM's bulletin for the CVE list, apply them in the next window, and treat SSH, SFTP and every other path through 5733-SC1 as part of the door they protect.

IT Jungle · Nov 2, 2026Approved · Reggie

FRIDAY · OCT 30

02
Govern02 The doorCommerce

Shopify puts card-network agent payments into Shop Pay, and the agent never sees the card

Shopify has built Visa, Mastercard and American Express agentic payment technology into Shop Pay, so a buyer's agent can check out at merchants on or off Shopify without ever holding the card. If agent orders reach your IBM i order entry through a storefront, record them as agent-placed with the network token reference, and test that payment matching and chargeback handling accept a token where a card used to be.

PYMNTS · Oct 29, 2026Approved · Allen

THURSDAY · OCT 29

02
Adopt02 The doorIBM i

Fall TRs require sign-on for telnet printer sessions

The 7.6 TR3 and 7.5 TR9 enhancements add authentication to telnet printer sessions: a user name and password or Kerberos is now required, and a user can print only to devices and output queues they are authorized to. Find every printer and emulator that opens an unauthenticated telnet print session before the PTFs go on, give each one a named profile with only the output queue authority it needs, and read IBM's Memo to Users before you schedule the install.

05
Govern05 The recordIBM i

IBM i adds SQL services for tracking CVE currency

The 7.5 TR9 and 7.6 TR3 enhancements list new SYSTOOLS services, CVE_CURRENCY and CVE_PTFS, beside a CVE Currency table in Navigator for i and new audit journal services for Directory Server and network authentication. Once they are on, schedule a query against those services and keep its output as part of your security record, so which CVEs a partition still carries is answered from Db2 for i, not from a spreadsheet.

03
Build03 Agents proposeCommerce

Oracle's new supply chain agents prepare plans and wait for approval

Oracle's new Fusion supply chain agents analyze spend and scarce supply, then prepare allocation plans and corrective actions for a team to review and approve, and the one agent that acts only executes plans already authorized. If your planning and purchasing run on IBM i, build agents the same way: they write proposals to a staging file, and only a named buyer's approval releases a PO or an allocation into the live files.

Oracle · Oct 26, 2026Approved · Allen

WEDNESDAY · OCT 28

03
Govern03 Agents proposeCommerce

Walmart lets shoppers cap agent spend, with approval on by default

Walmart's new Agent Profiles let a shopper authorize an agent and set a spending cap, with per-purchase approval as the default, while an MCP-based Agent Gateway in limited beta lets agents search, cart and buy without ever seeing card details. If your IBM i order entry will take agent orders, build the same split now: the agent proposes the cart through a scoped API, and a named person's approval or a recorded cap commits it to the order file.

BXTData · Oct 28, 2026Approved · Reggie
02
Govern02 The doorIBM i

IBM gives AI agents their own identity record and short-lived credentials

IBM made Agent Identity and Vault 2.2 generally available at TechXchange, so each agent gets an owner, a lifecycle and scoped, short-lived credentials that can be revoked the moment it steps out of bounds. IBM i shops letting agents reach Db2 for i should stop handing them a shared user profile: give each agent its own profile with least authority, expiring credentials, and a named owner who can switch it off.

01
Adopt01 Bob buildsIBM i

IBM i 7.6 TR3 and 7.5 TR9 are announced, shipping November 20

IBM announced the fall Technology Refreshes at TechXchange for November 20, adding header options for decompression and logging to the QSYS2 HTTP functions, plus RPG list initialization of arrays, a new %INFO built-in and a compile option that stops UPDPGM from changing a program. Plan the install on a test partition first, and use the UPDPGM lock on programs you want frozen once a person has reviewed them.

RPGPGM.COM · Oct 28, 2026Approved · Reggie

TUESDAY · OCT 27

05
Build05 The recordIBM i

Time every IBM i session from the history log with one SQL query

IT Jungle's Guru column pairs job-start CPF1124 and job-end CPF1164 messages from QSYS2.HISTORY_LOG_INFO on job number and user, then uses TIMESTAMPDIFF to show how long each session ran. Point the same query at the profiles your agents and service jobs run under, and land the rows in a table you keep, because the history log only reaches back as far as your cleanup settings allow.

IT Jungle · Oct 26, 2026Approved · Allen

MONDAY · OCT 26

02
Adopt02 The doorIBM i

PHP2Shell flaws reach IBM i web servers; patched PHP builds are out

Two critical PHP flaws let a single unauthenticated HTTP request run code on PHP 5.4 through 8.5, and the IBM i PHP distributions shipped fixes from October 21, including for some end-of-life releases. Inventory every PHP instance on the box, patch it from its distribution this week, and treat any web-facing PHP job as a door that may already be open: check its logs and the IFS for changes.

02
Adopt02 The doorIBM i

This week's IBM i PTFs close auth holes in SST/DST and Navigator, and block old ACS drivers

IBM's latest security bulletins for 7.3 through 7.6 fix improper authentication in SST/DST and Navigator for i, plus flaws in the HTTP Server admin GUI, DCM and zlib, and one PTF blocks connections from older ACS Windows drivers. Apply the PTFs, but upgrade ACS on every PC first so the new block doesn't cut off users still on old drivers.

FRIDAY · OCT 23

03
Govern03 Agents proposeCommerce

Stripe lets agents move company money, with human approval as a setting

Stripe Treasury for agents lets an agent pay invoices and move funds over Stripe's MCP server, with read-or-act scope and human approval set per business, and spending limits still weeks out. If an agent will ever propose payments against your IBM i payables, keep the approval gate and the record of who approved in Db2 for i, not only in the provider's console.

04
Govern04 A person commitsCommerce

Local payment schemes form a council to set agent checkout rules

PPRO has convened BLIK, EPI, Bancontact and DIAS to write shared standards for agents paying over non-card rails, covering out-of-band authorisation, agent attestation, mandates and delegated authority, and disputes. Card networks will not be the only place consent gets defined, so model the mandate as data your IBM i order system holds and checks, whatever rail carries the payment.

THURSDAY · OCT 22

02
Adopt02 The doorIBM i

ACS 1.1.9.17 patches CVEs in its SFTP client

IBM shipped Access Client Solutions 1.1.9.17 with fixes for CVEs in the SFTP client and an SQL error on RESTRICT ON DROP. ACS is a door into the box on every desktop that runs it, so roll the new build out like a PTF, retire the stray older copies, and read IBM's ACS updates page for the CVE detail.

WEDNESDAY · OCT 21

04
Govern04 A person commitsCommerce

Visa, BILL and Billtrust pilot agents paying buyer-approved B2B invoices

A live pilot lets an AP agent pay a buyer-approved invoice on a supplier portal, with Visa's Trusted Agent Protocol verifying the agent and a token standing in for the card. The agent never chooses what to pay, and an IBM i shop running AP on the core should keep that same split: the approval is written in the ERP, and the agent only executes it.

05
Govern05 The recordCommerce

Mastercard's Agent Pay for Machines caps agent spend with signed vouchers

Mastercard is rolling out Agent Pay for Machines on the x402 protocol: an agent that gets a 402 Payment Required signs a voucher carrying a human-approved spending limit, and the provider redeems it later. That is authority kept as a record, not a login, so any shop letting agents buy data or compute should set a per-agent ceiling by a person and reconcile every voucher back to it.

TUESDAY · OCT 20

02
Govern02 The doorCommerce

Know Your Agent is still competing schemes, not one standard

Heading into Money20/20, the card networks, Google's AP2 (now at the FIDO Alliance) and an IETF draft each prove an agent's identity and authority a different way, and none has become the standard. Shops opening order entry or catalog APIs to agents should verify agents at one gateway in front of IBM i, so the scheme can change without touching the RPG behind it.

MONDAY · OCT 19

05
Govern05 The recordCommerce

Agent checkout reaches TV and lock-screen shopping surfaces

PayPal is now the native checkout inside Glance's AI shopping surfaces on smart TVs and phone lock screens, and nobody has yet settled who pays when an agent buys the wrong thing. If IBM i is your order system of record, flag agent-originated orders and keep the buyer's mandate with the order before the returns and disputes arrive.

FRIDAY · OCT 16

02
Govern02 The doorCommerce

Williams-Sonoma's new terms bar AI shopping agents without written consent

Williams-Sonoma rewrote its terms so no agent may shop or use a customer account without the company's written permission, even when the customer authorized it, and every automated tool must identify itself. If you sell online from IBM i, set your agent policy at the door now, in your terms and in code, and log which agents you let through to an order.

KQED · Oct 13, 2026Approved · Reggie

THURSDAY · OCT 15

05
Adopt05 The recordCommerce

Amex publishes an open spec that binds agent purchases to approved intent

American Express released its Intent Intelligence Framework: the shopper approves a scoped intent, the agent gets a payment token tied to it, and off-intent purchases can be declined before they clear. If your order flow runs on IBM i, plan to store the intent ID with the order in Db2 for i, because it is the evidence that a person committed to that purchase.

TechCrunch · Oct 13, 2026Approved · Reggie

Built In.

The best of the log, every two weeks. One idea, one proof, one next step.

References do not imply endorsement. Links go to the original source; the words are ours. Alder & Pine Home and Silver Lake Financial Group are fictitious demonstration companies.