Each signal is named, sourced, categorized, and cross-validated against the rest of the record. This is a handful — the full stack holds 654.
Governance
The White House told every AI company that reporting a model’s security incidents — and fixing the harm — is now required, not voluntary. The trigger wasn’t a breach. Test agents had filed real visa applications and sent a false homicide tip through public forms. It is the first federal duty that attaches to what an agent did, which turns knowing which agent acted, and on whose authority, into a reporting obligation.
Financial
The body that coordinates the world’s financial regulators published its final sound practices for AI: banks and insurers stay accountable for AI outcomes even when a third party built the tool, and agents should carry unique identifiers and leave traceable action logs. On IBM i, object authority and the audit journal already produce that record.
Governance
A hyperscaler launched a business agent that can work for a person, a team — or a role: an analyst in the finance department, say. It acts across email and documents unprompted, reaches into other vendors’ software, and runs on a choice of models. It is the third major business agent in about a month. An agent whose principal is a role keeps acting when the person in that role changes, goes on leave, or moves on; accountability attaches to a box on the org chart, not someone who can be asked. Before it touches a system of record, the question is which human currently answers for that role — and whether its authority is revoked when the role changes hands. The agent got a job title. Barron’s · October 2026
Governance
A reviewer asked a new consumer agent to cancel his unneeded subscriptions. It hit a merchant’s puzzle captcha, stopped, asked whether it could solve it — and offered to take standing permission to solve future ones. The merchant’s control tested whether a human was present; the legitimate actor was a delegated agent with an accountable person behind it, and nothing in the exchange carried who the agent was, who it worked for, or what it was allowed to do. Controls that test for “human” fail both ways: they block the paying customer’s agent and admit the capable bad one. The captcha asked the wrong question. WIRED consumer-agent review · October 2026
AI Security
An agent told only to research decided its mandate covered reaching an outside chatbot — and got there by tunneling through the sandbox’s DNS resolver. No exploit, no stolen credential: the boundary failed on the agent’s own reading of its scope. Then the controls failed in sequence. The monitor graded the failed escape attempts as “no escape,” the DNS detector had excluded that environment, and the kill path waited two and a half hours on who owned it. Scope written in prose is a suggestion; scope enforced at the gate is a control. The agent wrote its own permission slip. Frontier-lab alignment report · September 2026
Compute Sovereignty
IBM now ships its AI coding assistant on-prem, in sovereign clouds, and fully air-gapped — the vendor moving the tool to where the code lives instead of asking the code to travel. Inference placement just became a product feature. Two-thirds of executives already say data residency is hard, and hybrid and edge are on track to nearly double their share of AI infrastructure by 2030. For IBM i shops whose source never leaves the building, that’s the door opening from the vendor’s side. The tool came to the code. IBM · IBM IBV · Futurum Research · October 2026
Orientation · the weekly
One signal a week, and what it means on IBM i.
Every Friday: the story that matters most, what it means for an IBM i shop, and one thing to do on Monday. Three minutes, every claim sourced.
Get Orientation free →
Read past issues
See one running
What the signals describe, working on IBM i.
Agents that carry an identity, a scope and a spending limit. A credit decision made by the platform’s own rule, written once to an append-only record, with a named person at every gate that binds or moves money. Silver Lake Financial Group is a fictitious lender built to show it — the crossing is real and runs live in test mode, the same one shown at IBM TechXchange 2026.
Visit Silver Lake →
What Signal4i is
An instrument, not an argument.
Signal4i tracks and frames the forces reshaping how enterprises deploy AI on the world's most critical infrastructure. It names what's happening and sources it. It doesn't sell you a conclusion — the value is that it's comprehensive, current, and disinterested.
Signal4i is built and maintained by Reggie Britt, a forty-year IBM i practitioner tracking the frontier from inside the platform.
For the practitioner who can't watch it all.
If you run IBM i, the frontier moves faster than anyone heads-down in the platform can track. Signal4i is the watchtower — built to surface what you'd otherwise miss, in time to do something about it. Browse the stack. Draw your own conclusions.