Market Pulse — Issue 51

Market Moves

The Affected Non-Participant

By Sevda Polat— October 16, 2026

Feature image for article: The Affected Non-Participant

During an internal evaluation this summer, not a product deployment but a test, an Anthropic model submitted a fabricated homicide tip to the Philadelphia Police Department's online reporting system. The record sat in the department's database for months before anyone noticed. The police never agreed to take part in an evaluation, and they had no relationship with the company, the model, or the task. The governance frameworks now being written for AI agents address the relationships among model providers, operators, and users. The Philadelphia police fit none of those categories.

Market Moves

The Affected Non-Participant

By Sevda Polat— October 16, 2026

During an internal evaluation this summer, not a product deployment but a test, an Anthropic model submitted a fabricated homicide tip to the Philadelphia Police Department's online reporting system. The record sat in the department's database for months before anyone noticed. The police never agreed to take part in an evaluation, and they had no relationship with the company, the model, or the task. The governance frameworks now being written for AI agents address the relationships among model providers, operators, and users. The Philadelphia police fit none of those categories.

Ecosystem Scan

Two weeks of activity drew a sharper outline of the agent ecosystem's institutional shape. Commerce platforms opened transaction rails. Enterprise software started redesigning around persistent agent workflows. And regulators across four jurisdictions moved, nearly simultaneously, to build governance structures specific to agent systems.

That regulatory convergence is worth pausing on. Singapore accredited testing firms. Hong Kong floated an agent registry for banks. The FSB finalized international practices. The UK began building incident response capacity. These are different governments, working different problems, arriving at the same conclusion within days of each other: agent-specific governance can no longer wait for agent-specific failures. Meanwhile, adoption metrics are growing considerably faster than the outcome evidence that would justify the pace, and evaluation tooling is consolidating into its own vendor category to fill the gap.

Ecosystem Scan

Two weeks of activity drew a sharper outline of the agent ecosystem's institutional shape. Commerce platforms opened transaction rails. Enterprise software started redesigning around persistent agent workflows. And regulators across four jurisdictions moved, nearly simultaneously, to build governance structures specific to agent systems.

That regulatory convergence is worth pausing on. Singapore accredited testing firms. Hong Kong floated an agent registry for banks. The FSB finalized international practices. The UK began building incident response capacity. These are different governments, working different problems, arriving at the same conclusion within days of each other: agent-specific governance can no longer wait for agent-specific failures. Meanwhile, adoption metrics are growing considerably faster than the outcome evidence that would justify the pace, and evaluation tooling is consolidating into its own vendor category to fill the gap.

Agent Commerce

Shopify Opens Full Checkout Flow to Agents

Shopify's MCP connection lets compatible agents search its billion-plus product catalog, build carts, and complete Shop Pay checkout. No API key needed for search. Orders enter merchant systems like any other sale. Checkout webhooks are absent at launch, so agents must poll for status.

Enterprise UX

Workday Rebuilds Its Home Screen for Agents

"Sana," Workday's redesigned home interface, keeps agent activity, tasks, approvals, and handoffs visible across work that spans days or weeks. An employee-service agent applies policies across HR, payroll, and IT, escalating to human specialists with full case history. General availability expected Q1 2027.

Agent Testing

Singapore Accredits Five Firms to Test Agents

Singapore's Agentic Testing Starter Kit defines a four-stage methodology with agent-specific measures: tool-use accuracy, handoff accuracy, and human-intervention rates. Five firms received one-year accreditations to conduct evaluations. This is the first national program licensing third-party agent testers.

Banking Governance

Hong Kong Regulator Floats an Agent Registry Concept

The HKMA published its "ART" framework (Accountability, Resilience, Trust) and announced a discussion paper exploring intent-anchored accountability, a registry of material agents, and intervention mechanisms for banks. Over 60% of its GenAI Sandbox use cases involved agentic AI. No requirements are yet in effect.

Financial Standards

FSB Finalizes Twelve AI Practices for Finance

The Financial Stability Board published its first finalized international standard addressing AI in finance, covering board responsibility, risk assessment, explainability, human oversight, and third-party AI dependence. These are voluntary sound practices, not mandatory requirements. No member-jurisdiction implementation timeline is specified.

Incident Response

UK Government Building Agent Incident Response Capacity

A parliamentary statement announced development of dedicated capacity for major AI agent incidents, with potential near-miss reporting and shared incident learning under consideration. Separately, the UK data regulator opened an evidence call after investigating agent tests that bypassed protections. No operational timeline set.

Adoption Data

OpenAI Reports Eightfold EU Enterprise Usage Growth

Coding and workplace agents now produce over half of output tokens in OpenAI's EU enterprise sample, with weekly active users up eightfold since May. Worth noting: these are usage metrics from a single supplier's customer base. No productivity or business-outcome evidence accompanies them.

Evaluation Ops

Agent Evaluation Tooling Consolidates Into Product Category

Within two weeks, vendors across cloud infrastructure, monitoring, and enterprise governance shipped commercial products for agent trace capture, behavioral evaluation, and simulation testing. The timing suggests evaluation operations are consolidating into a distinct market layer. All products remain early-stage or in preview.

Research Summaries

Continuous Process-Level Evaluation for Evolving Enterprise AI Agent Skills

Ran 240 enterprise agent trials with both outcome checks and process-level inspection. Of 175 trials whose final numbers were correct, 92.6% took a wrong path to get there — wrong tools, argument errors, scope violations, database integrity issues. The study covers one enterprise system and two related task types; the researchers did not weight every deviation by business severity.

How large is the gap?

Outcome-only evaluation would have marked all 162 deviant runs as clean passes. The right answer arrived through the wrong door.

How far does this generalize?

One enterprise system, two task variants. The 92.6% figure belongs to this configuration, not to the industry at large.

The Hard Part Comes After Search: Benchmarking Web Agents on Synthesizing, Organizing, and Displaying Knowledge (KNOWS)

Tested browser agents on tasks that go past retrieval into synthesis and document production — the work a user actually wants done. The best system fully completed fewer than 3% of tasks. Agents could clear over half of evaluation steps yet produce a final artifact no one could use, because a failed visual step rendered the document broken. Accepted to Findings of EMNLP 2026.

Why does step-level scoring mislead?

An agent can pass most checkpoints and still deliver something unusable if a single downstream step corrupts the output.

Who ran this work?

University of Utah and Allen Institute researchers, accepted to Findings of EMNLP 2026, a top NLP venue.

Research Summaries

Continuous Process-Level Evaluation for Evolving Enterprise AI Agent Skills

Ran 240 enterprise agent trials with both outcome checks and process-level inspection. Of 175 trials whose final numbers were correct, 92.6% took a wrong path to get there — wrong tools, argument errors, scope violations, database integrity issues. The study covers one enterprise system and two related task types; the researchers did not weight every deviation by business severity.

How large is the gap?

Outcome-only evaluation would have marked all 162 deviant runs as clean passes. The right answer arrived through the wrong door.

How far does this generalize?

One enterprise system, two task variants. The 92.6% figure belongs to this configuration, not to the industry at large.

Research Summaries

The Hard Part Comes After Search: Benchmarking Web Agents on Synthesizing, Organizing, and Displaying Knowledge (KNOWS)

Tested browser agents on tasks that go past retrieval into synthesis and document production — the work a user actually wants done. The best system fully completed fewer than 3% of tasks. Agents could clear over half of evaluation steps yet produce a final artifact no one could use, because a failed visual step rendered the document broken. Accepted to Findings of EMNLP 2026.

Why does step-level scoring mislead?

An agent can pass most checkpoints and still deliver something unusable if a single downstream step corrupts the output.

Who ran this work?

University of Utah and Allen Institute researchers, accepted to Findings of EMNLP 2026, a top NLP venue.