Microsoft fields 100+ collaborating AI agents to hunt AI-driven cyberattacks

What changed: Microsoft unveiled Perception, a cybersecurity platform where more than 100 specialized AI agents work in coordinated Red, Blue, and Green teams to simulate attacks, detect threats, and automatically block vulnerabilities. Perception combines these agents with a new security-optimized model, MAI-Cyber-1 Flash, and beat competing systems by roughly 10 percentage points in the CyberZoo benchmark.

Why it matters: Security and IT leaders can now treat autonomous agents as a permanent, multi-model defense layer rather than one-off scripts or rules, especially against machine-speed attacks. For builders of agentic systems, the design shows how to split roles into narrowly focused agents that exchange signals instead of relying on one general model.

Try/watch: Map your current security stack into Red, Blue, and Green-style roles and identify where specialized AI agents could take over repetitive detection and triage, while monitoring how Perception’s benchmarks evolve against real-world attacks.

Fujitsu builds multi-agent financial platform with sovereign data and trusted AI

What changed: Fujitsu announced development of the Uvance for Finance AI Transformation Platform, built on its Kozuchi Enterprise AI Factory to guarantee data sovereignty and trustworthy AI for regional banks and other financial institutions. The platform will use Fujitsu’s Takane language model and multiple specialized AI agents, such as customer understanding and sales optimization agents, to support loan screening, inquiries, document creation, and eventually branch operations and compliance when it launches in 2027.

Why it matters: Financial operators get a roadmap for deploying agents inside strict regulatory and data-control boundaries, rather than relying on generic cloud AI. For founders, the design reinforces that multi-agent systems plus domain-specific models can unlock higher-value workflows in finance where generic chatbots have struggled.

Try/watch: Review where your financial or fintech products touch regulated data and consider how sovereign hosting, domain models, and task-specific agents could be bundled into a single platform offering like Fujitsu’s.

Emerson rolls out Ovation AI agents to automate power and water plant operations

What changed: Emerson expanded its Ovation Automation Platform 4.0 with five new AI agents for power and water plants: Sequence Assistant, Alarm Insight, Predictive Maintenance, Root Cause, and Loop Performance Monitor agents. These agents continuously monitor control logic, alarm floods, equipment health, process anomalies, and control loop performance, then generate evidence-backed recommendations to help operators prioritize actions and prevent failures.

Why it matters: Operators of critical infrastructure can move from static rules and dashboards to always-on agents that watch for deviations, explain them, and propose interventions before outages or safety incidents. For industrial AI builders, Emerson’s launch illustrates how packaging narrow, high-impact agents by operational role can make adoption easier than selling a single monolithic “plant brain.”

Try/watch: Identify one high-friction operational problem—such as alarm overload or unplanned downtime—and scope a single agent around it, mirroring Emerson’s focused agent patterns before attempting broader automation.

Schrödinger’s Bunsen positions agentic AI as a lab co-scientist

What changed: Schrödinger introduced Bunsen, an early-access agentic AI co-scientist that helps researchers define scientific objectives, design computational strategies, run complex molecular discovery workflows, and interpret the results. The product is available initially to select customers, with full commercial release targeted by the end of 2026.

Why it matters: R&D teams in pharma and materials can treat AI agents not just as analysis tools, but as collaborators that plan and execute multi-step experiments across simulations and data pipelines. This raises the bar for scientific software vendors, who will need to embed agentic orchestration and explanation into their platforms to stay competitive.

Try/watch: Audit your current discovery workflows to pinpoint where scientists manually coordinate tools or simulations, and design a pilot where an agent like Bunsen owns orchestration and reporting while humans focus on hypothesis and decision-making.