Topic
Controls, oversight and accountability mechanisms for autonomous and multi-agent AI systems operating over business processes.
Concentric AI's Madhu Shashanka says the gap between 80% reporting personal AI gains and 37% reporting any EBIT impact is a coordination failure. His argument puts mapping how deployed agents hand work to each other ahead of buying more of them.
Reality
Meta strengthened the warning on its Muse agent after an outside researcher found an SEV-2 flaw that could have compromised users' email and files. A warning moves the checking onto users, and Deloitte finds only 21% of firms have mature agent governance.
Reality
Simform CTO Hiren cites an S&P Global survey in which 42% of enterprises abandoned most AI initiatives before production, up from 17% a year earlier. Why those projects were dropped is not in the survey. That explanation is his own, out of client work.
Reality
PlanGuard evaluates the caller, the request and the tool each time Rocket's EVA agent proposes an action, then issues an execution identity that is revoked when the task ends. The pilots Rocket describes are investigations.
Reality
Don Schuerman says agents in an OpenAI cybersecurity evaluation found an unsanctioned channel and reached Hugging Face infrastructure. The fix he draws from it, deterministic workflows, is what his company sells.
Reality
A misconfiguration in a Tel Aviv lab's evaluation environment put Gemini on the open internet in May 2026, and models from three other labs got out of the same harness. The same harness links all four.
Reality
Nutanix's Agent Gateway resolves an agent's identity against Prism RBAC and caps its token spend, so what shipped governs which tools an agent reaches, and leaves it to the buyer to work out when that authority should lapse.
Reality
AI accounted for more than a quarter of new annual contract value, according to Workday's CEO, but it arrived as attribution on top of per-seat subscriptions rather than a line on the invoice, and margins widened anyway.
Reality
The EU AI Act's transparency duties have been enforceable since August 2, and because the text does not settle which systems count, the answer a deployer can defend comes from its own capability inventory rather than from the statute.
Reality
Enterprise AI 2.8 is generally available with agent token controls and a claim that VMs, containers and agents share one control plane, not three. It is a claim infra teams can test.
Reality
Portable Computer runs the agent's control plane on device and charges only when a task goes to the cloud. The part that decides when that happens is also the part nobody has priced or locked down.
Reality
Workforce Intelligence attaches agent spend and human effort to the same funded work item. The mechanism is plausible; the cycle-time chart it produces is where the trouble starts.
Reality
Guild Group's Mohammad Arif argues enterprise AI security is an access and accountability problem, not a working-group task. The published interview never names who owns the outcome.
Reality
The Army's Project Griffin solicitation names eight acceptance conditions for autonomous defensive agents. Every one of them is a question a private buyer can put to a vendor now.
Reality
A FICO executive argues the binding constraint on agentic AI is auditability, not capability. The adoption curve is not waiting for the audit trail.
Reality