All articles
AI News

AI News Roundup: April 8, 2026 — Agents Go Mainstream, Security Scrambles to Catch Up

JTJennifer T.R.Editor in Chief, Stronk Blog8 April 202610 min read

Today's AI landscape is moving at a pace that makes last month feel like ancient history. Here is everything that matters from April 8, 2026 — and what it means for businesses considering AI agents.

MYOB and Microsoft Partner to Bring AI Agents to Australian Small Business

This is a big one for anyone running a business in Australia. MYOB, the accounting software used by over 1.2 million Australian and New Zealand businesses, has signed a five-year strategic partnership with Microsoft to deploy AI agents directly inside MYOB's small business tools.

The agents will be built on Microsoft's Agent 365 governance platform and will be able to:

Forecast cash flow based on historical patterns and upcoming invoices
Guide compliance by flagging BAS deadlines, superannuation obligations, and regulatory changes
Handle customer support triage by resolving common queries before they reach your team
Generate case summaries for bookkeepers and accountants managing multiple clients

What makes this significant is the deployment speed. MYOB says the partnership has accelerated feature releases from months to weeks. For the million-plus businesses already using MYOB, AI agent capabilities will arrive through software updates rather than requiring a separate tool or migration.

This is a strong signal that AI agents are no longer the domain of enterprise tech companies. They are coming to the tools small businesses already use, built into the workflows that already exist.

96% of Organisations Now Use AI Agents — But Almost Nobody Has Governance

Here is the stat that should make every business owner pause: according to new enterprise survey data reported this week, 96% of organisations are now using AI agents in some capacity. That is near-universal adoption.

But the second stat is the concerning one: 94% of those organisations worry about uncontrolled agent sprawl. That means agents are being deployed faster than governance frameworks can manage them.

The core problem is that AI agents, unlike traditional software, can act autonomously. They can access APIs, read databases, send emails, and trigger workflows without a human in the loop for every action. When an organisation deploys dozens of these agents across departments without a centralised understanding of what each one does, what data it accesses, and what actions it can take — the risk profile grows exponentially.

What this means for small business

You probably do not have dozens of agents yet. But even with one or two, the principle applies. Know exactly what your agent can access, what actions it can take autonomously versus what requires your approval, and have a clear process for reviewing its activity. This is not about slowing down — it is about deploying responsibly so you do not create problems that cost more to fix than the agent saves.

Automation Anywhere: AI Agents Now Auto-Resolve 80% of IT Support Requests

Automation Anywhere published data this week showing that AI agents deployed in IT service management environments are now auto-resolving over 80% of support requests without human intervention, according to industry reporting from AI Agent Store.

The headline numbers:

80%+ auto-resolution rate for common IT tickets (password resets, access requests, software installation, troubleshooting guides)
Up to 50% cost reduction in IT service management
Potential savings exceeding $5 million annually for large enterprises
Deployment timeline: 8 weeks from decision to live

This mirrors what we see across other industries. The tasks that AI agents handle best are the ones that are high-volume, follow predictable patterns, and currently consume disproportionate human time. IT support tickets fit that profile perfectly — and so do phone calls, appointment bookings, invoice chasing, and data entry.

The 8-week deployment timeline is particularly notable. That is faster than most traditional software implementations and suggests the tooling has matured to the point where deployment is no longer the bottleneck — organisational readiness is.

AudAgent: New Research Shows AI Agents Failing Basic Privacy Tests

Researchers at the Rochester Institute of Technology released AudAgent, a monitoring tool designed to audit how AI agents handle sensitive data. The results are sobering.

In testing, agents powered by Claude, Gemini, and DeepSeek failed to refuse requests for Social Security numbers when the requests were embedded in multi-step workflows. GPT-4o performed better but was not immune. The research found that agents risk storing and sharing:

Passwords and authentication credentials
Health and medical data
Location information
Financial details

This does not mean AI agents are inherently unsafe. It means that the default behaviour of many AI models is not calibrated for the sensitivity levels that business data requires. Agents need to be configured with explicit rules about what data they can and cannot access, store, or transmit.

The takeaway for business owners

If you are deploying an AI agent that touches customer data, financial records, or any information covered by privacy legislation (and in Australia, the Privacy Act 1988 covers most business data), you need to understand exactly how your agent handles that data. Where is it processed? Is it sent to a cloud provider? Is it stored locally? Can it be included in model training data?

These are not hypothetical questions. They are compliance requirements.

Anthropic Captures 40% of Enterprise LLM API Spend — OpenAI Drops to 27%

The balance of power in the AI model market has shifted significantly. According to market data reported this week, Anthropic now captures 40% of enterprise LLM API spending, up from a much smaller share just two years ago. OpenAI has declined to 27%, down from roughly 50% in 2023.

The drivers behind this shift:

Claude's reliability — enterprise customers report fewer unexpected outputs and more consistent performance in professional workflows
Anthropic's safety-first positioning — in regulated industries like finance, legal, and healthcare, the emphasis on safety and controllability resonates with compliance teams
The Model Context Protocol (MCP)Anthropic's open protocol crossed 97 million installs in March 2026 and has become foundational infrastructure for building AI agents, with every major provider now shipping MCP-compatible tooling
Price competition — Anthropic has been aggressive on pricing for high-volume enterprise contracts

For businesses choosing which AI model to run their agents on, this market shift matters. It means you have genuine choice, prices are falling, and you are not locked into any single provider.

Microsoft Releases Agent Framework 1.0.0

Microsoft released version 1.0.0 of its Agent Framework this week — a significant milestone that signals the tooling for building AI agents has reached production maturity.

The framework fundamentally rethinks how developers build agents by separating agent control from applications. In practical terms, this means:

Agents can be connected to pre-configured systems in Azure without custom integration code
The framework handles authentication, data access, and workflow orchestration
Developers focus on defining what the agent should do, not how it should connect to infrastructure

This is relevant for businesses because it lowers the barrier to building custom agents. What previously required weeks of development can now be configured in days. The framework also includes governance features — audit trails, permission boundaries, and activity monitoring — that address the sprawl concerns mentioned above.

The Projection: 40% of Business Applications Will Use AI Agents by End of 2026

Multiple analyst reports converged on the same figure this week: by the end of 2026, 40% of business applications will employ AI agents, up from under 5% in 2025. That is a roughly 8x increase in a single year, according to research compiled by industry analysts.

The applications leading adoption:

IT service management (as the Automation Anywhere data above shows)
Customer service (call handling, ticket routing, first-response automation)
Finance and accounting (transaction reconciliation, report generation, compliance checks)
HR (candidate screening, onboarding workflows, policy queries)
Sales (lead qualification, CRM updates, follow-up sequences)

The gap between early adopters and the rest is widening. Businesses that deploy agents now are building institutional knowledge — their agents are learning their workflows, accumulating business-specific data, and improving. That advantage compounds over time.

What This All Means

The theme of today's news is clear: AI agents are no longer experimental. They are in production at 96% of organisations. They are being embedded into mainstream business tools like MYOB. They are auto-resolving 80% of IT tickets. And the tooling to build, deploy, and govern them has reached 1.0 maturity.

But the governance gap is real. Agents are being deployed faster than organisations can track them. Privacy and security testing shows that default configurations are not always safe. And the shift in market share between AI providers means the landscape is still fluid.

For business owners, the practical advice has not changed: start with a specific problem, deploy carefully, understand what your agent can access, and measure the results. The technology is ready. The question is whether your implementation is thoughtful enough to capture the value without creating new risks.

---

*Stronk AI builds and deploys custom AI agents for Australian businesses. If you want to understand how agents apply to your specific workflows, book a free consultation.*

Discussion

Ready to put this into practice?

Book a free consultation and we will show you exactly how an AI agent applies to your business.

Book free consultation