Enterprise AI Brief — 2026-09-28

Today: Enterprise AI is moving decisively from experimentation toward governed execution, with agent security, runtime controls, and production-grade workflows emerging as the critical infrastructure layer.

Top Stories

1. 🔒 NVIDIA launches Open Agent Safety Platform for governed AI agents

NVIDIA · 2026-09-28

Bottom line: NVIDIA introduced an open agent-safety architecture combining OpenShell runtime controls with an out-of-band Sentry design to constrain and monitor autonomous AI agents.

The platform is designed to provide governance from agent testing through deployment, rather than relying solely on model-level safeguards. OpenShell establishes a controlled runtime for agent actions, while Sentry uses separate infrastructure to monitor behavior and support rapid quarantine when agents move outside defined boundaries.

Why it matters: Enterprise agents increasingly have access to credentials, data, networks and business tools; security therefore has to move below the prompt and model layer into runtime and infrastructure enforcement.

🔗 Read the full story

2. 🔒 Thales expands Google Cloud partnership to secure agentic AI workflows

Thales · 2026-09-28

Bottom line: Thales and Google Cloud are integrating security controls across interactions between users, AI agents, models, enterprise data and tools.

The expanded collaboration connects Thales AI Security Fabric with Google Cloud Gemini Enterprise to provide visibility, policy enforcement and protection for agent-driven workflows. The companies are targeting controls over what agents can access, share and execute as enterprises move beyond conversational assistants.

Why it matters: Agent governance is becoming an enterprise control-plane problem spanning identity, data, models and tools rather than an isolated AI-model security problem.

🔗 Read the full story

3. 🔒 Noma extends AI-agent security and governance to employee endpoints

Noma Security · 2026-09-28

Bottom line: Noma introduced endpoint controls that discover, govern and protect AI agents, MCP servers and agent skills running on employee machines.

The new capabilities provide discovery and access control for agents operating on laptops and add runtime detection and response intended to block dangerous behavior. Noma says policies can follow agents across SaaS, internally developed and endpoint environments through a common control plane.

Why it matters: As employees install and operate increasingly capable agents locally, enterprise AI governance must extend beyond centrally managed cloud applications to the endpoint and its existing user permissions.

🔗 Read the full story

4. 🤖 dv01 launches agentic infrastructure for structured finance

dv01 · 2026-09-28

Bottom line: dv01 launched AI agents for structured-finance workflows alongside MCP connectivity that lets client-built AI applications use its loan data and analytics.

The new capabilities cover DealStudio and Credit Facility Management, with agents able to work from source documents and transaction data. Its MCP server allows external AI applications to access collateral analysis and cash-flow modeling, while a semantic layer supplies standardized financial meaning and methodology.

Why it matters: The deployment illustrates an important enterprise-AI pattern: domain-specific semantic layers and controlled tool access can be more important than the underlying model when AI enters regulated, data-intensive workflows.

🔗 Read the full story

5. 💳 GoComet launches AI execution layer for logistics workflows

SplashTech · 2026-09-28

Bottom line: Singapore-based GoComet launched Nova, an AI-native execution layer that lets agents take defined actions across freight planning, procurement, compliance, shipment and payment workflows.

The platform connects enterprise data, workflow rules, external logistics partners and AI agents rather than positioning AI as a standalone chatbot. Demonstrated workflows included identifying document errors, managing service tickets, comparing freight quotes and progressing shipment bookings subject to configured limits and human approval.

Why it matters: Enterprise AI is shifting from information retrieval toward controlled execution, with workflow permissions and approval thresholds becoming core design primitives.

🔗 Read the full story

6. 🤖 AWS Builder Studio highlights the shift from AI prototypes to production

Omdia · 2026-09-28

Bottom line: Omdia’s observations from AWS’s Melbourne Builder Studio point to governance, resilience, stakeholder alignment and measurable outcomes as the main barriers to scaling enterprise AI.

The report describes enterprises and partners using the studio model to validate AI ideas before production, including agentic foundations for financial institutions. It highlights production examples such as Westpac’s agentic AI deployment and emphasizes defining governance, observability, unit economics and human oversight before prototypes are built.

Why it matters: The enterprise AI bottleneck is increasingly moving away from model capability toward operating discipline, architecture, governance and measurable business value.

🔗 Read the full story