A new model every week. A new study every month proving that all the previous studies were wrong. And now: a compliance deadline that hits in 75 days.
This article is our monthly relevance filter for small and mid-sized businesses. We read the release notes, legislative drafts, and benchmark studies — and give you a straight answer: Which of these updates actually affects a company with 10 to 500 employees? What can wait? And what should already be on your desk this week?
May 2026 was a packed month. Google I/O, new Anthropic models, a serious agentic AI wave, and an EU AI Act countdown you can no longer afford to ignore.
Quick Scan: 8 Updates at a Glance — Filtered by Your Stack
| Update | Provider | Relevance | Priority |
|---|---|---|---|
| Claude Opus 4.7 & Sonnet 4.6 | Anthropic | Document analysis, agents | High |
| Claude for Small Business | Anthropic | HubSpot, QuickBooks, Workspace | High |
| GPT-5.5 in Microsoft 365 Copilot | OpenAI / Microsoft | Office users | High |
| Google I/O: Proactive AI | Workspace users | Medium | |
| Agentic AI (Orbit, Zapier) | Anthropic / Zapier | Automation teams | Medium |
| EU AI Act August deadline | EU | All companies using AI | Critical |
| AI & Cybersecurity (mAI 2026) | BSI / Transferstelle | All | High |
| Market data (Bitkom, Salesforce) | Research | For internal strategy discussions | Medium |
1. Claude Opus 4.7 & Sonnet 4.6: The Most Capable Model Is Now Generally Available
Anthropic launched two models simultaneously in May — and the timing is no coincidence. Claude Opus 4.7 is now the most powerful available model for complex tasks, especially for reasoning, coding, and multi-step AI agents. Claude Sonnet 4.6 outperforms even pricier competitor models on coding benchmarks — at the same price as its predecessor.
What’s actually changed:
- Extended agent capabilities: Opus 4.7 can execute complex multi-step tasks autonomously, using tools and making decisions without requiring human intervention at each step
- Improved instruction adherence: The model reliably stays within defined boundaries and formatting requirements — critical for production use
- Sonnet 4.6 as the cost-efficiency winner: For most business applications, Sonnet 4.6 is the better choice — more capable than many frontier models, at Sonnet pricing
Current API pricing (May 2026):
| Model | Input ($/MTok) | Output ($/MTok) | Batch API |
|---|---|---|---|
| Haiku 4.5 | $1 | $5 | 50% discount |
| Sonnet 4.6 | $3 | $15 | 50% discount |
| Opus 4.7 | $5 | $25 | 50% discount |
What this means for you:
If you’re already using Claude via the API or tools like n8n, upgrading to Sonnet 4.6 is in most cases a pure configuration change. If you’re planning complex document analysis, multi-step research workflows, or AI agents, Opus 4.7 is currently the most capable tool available.
For a direct comparison between Claude and ChatGPT for specific business applications, see our Claude vs. ChatGPT comparison.
2. Claude for Small Business: Ready-Made AI Workflows — Deployable Today
This is the update of the month that barely made headlines — but it’s directly relevant for small and mid-sized businesses: Anthropic launched “Claude for Small Business”, a set of pre-built AI integrations for the tools companies already use.
Currently available as native integrations:
- QuickBooks and PayPal: Automated bookkeeping, invoice processing, and payment reconciliation
- HubSpot: AI-assisted lead qualification, email sequences, and CRM data maintenance
- Canva: AI-powered creation of marketing materials based on brand guidelines
- DocuSign: Generate contract drafts, review clauses, prepare signature workflows
- Google Workspace and Microsoft 365: Deep integration across email, calendar, documents, and spreadsheets
Why this matters: Until now, integrating Claude into existing business processes required either development resources or middleware tools like n8n. Claude for Small Business makes this an out-of-the-box experience. A small business using HubSpot and QuickBooks can activate AI support in both systems within an hour — no IT contractor needed.
3. GPT-5.5 in Microsoft 365 Copilot: Now Handles Multi-Step Tasks
On April 27, 2026 — just four days after its introduction — GPT-5.5 was integrated into Microsoft 365 Copilot. What this means: Copilot can now autonomously execute multi-step tasks that previously required manual intervention at each stage.
Concrete new capabilities:
- Autonomous process execution: Copilot can now do more than summarize and draft — it can work through tasks sequentially across multiple apps. For example: take meeting notes from Teams, create tasks in Planner from those notes, and send a summary email
- Improved data analysis in Excel: Build financial models, variance analyses, and forecast scenarios using natural language prompts
- Deeper SharePoint integration: Use internal documents, policies, and manuals as knowledge sources for Copilot responses
The practical difference from GPT-5.2:
GPT-5.5 isn’t just “better” — it’s a qualitatively different tool. While GPT-5.2 handles individual tasks very well, GPT-5.5 can plan and execute processes independently. That’s the difference between a good tool and an assistant that actually thinks.
4. Google I/O: The Proactive AI Shift — What It Means for Your Daily Work
Google’s annual conference on May 15, 2026 had one central theme: the shift from reactive to proactive AI. Away from the chatbot that responds when you ask. Toward AI that surfaces relevant information before you even know to ask.
The most important announcements for businesses:
Google Workspace Intelligence (since April 22, Google Cloud Next ‘26): A new semantic layer across Gmail, Google Docs, Drive, Sheets, and Slides. Gemini now understands not just documents — it understands the user’s working patterns, recognizes context, and actively orchestrates tasks. Concretely: a proposal in Docs is recognized, the relevant CRM conversation from Gmail is automatically linked, and the next step is suggested.
Gemini Proactive Assistance (new at Google I/O): Gemini now proactively reaches out when it detects that information might be relevant to you — like an attentive assistant who shows you the important email before you have to search for it.
NotebookLM Enterprise (generally available): The business version of Google’s research tool lets you upload internal company documents — manuals, meeting notes, product documentation, training materials — as knowledge sources. Answers are no longer generic; they’re grounded in your company’s actual data.
What this means for Google Workspace users:
If you’re already on Google Workspace and haven’t activated the AI features yet, now is the time. The integration runs deeper here than with any other provider — and since February 5, 2026, there’s a new AI Expanded Access Add-On that bridges the gap between Standard and Enterprise tiers.
5. Agentic AI: The Mega-Trend Has Arrived in the Mid-Market
“Agentic AI” was for a long time buzzword territory reserved for tech circles. The Salesforce AI Index for Mid-Market 2026 (March 2026) turns it into hard reality: 16.6% of mid-sized businesses are already running AI agents — nearly double the figure from the previous year (8.7%).
What is an AI agent?
An AI agent is an AI system that autonomously pursues goals, uses tools, makes decisions, and plans multiple steps — without a human approving every individual action. The difference from a chatbot: a chatbot answers. An agent acts.
Two new developments in May 2026:
Anthropic Orbit (May 6, 2026): Proactive reporting — the AI agent reaches out on its own when it detects a relevant change or development. Example: the agent continuously monitors your order pipeline and alerts you when a threshold is crossed or an unusual pattern emerges.
Zapier AI Agents (officially out of beta): Autonomous agents for sales and marketing teams with no technical background. Particularly well-suited as an entry point if you don’t have an IT team in-house.
What mid-market companies are using in practice:
| Use Case | Estimated Time Savings | Barrier to Entry |
|---|---|---|
| Inbox triage & routing | 1–2 hrs/day | Low |
| Lead qualification & CRM updates | 30–60 min/day | Medium |
| Proposal prep from CRM data | 45–90 min/proposal | Medium |
| Invoice review & booking suggestions | 2–3 hrs/week | Medium–High |
| Supplier monitoring & alerting | Async | High |
6. EU AI Act: 75 Days to the August Deadline — and 78% Aren’t Ready
This is the only update in this article that isn’t optional.
On August 2, 2026, the full requirements of the EU AI Act for high-risk AI systems come into effect. Fines are no longer theoretical — they’re enforceable.
The current situation is concerning: According to a recent survey, 78% of SMEs are not yet audit-ready for EU AI Act compliance. These aren’t companies that don’t know what’s coming — these are companies that know and haven’t acted yet.
What many companies underestimate: You’re not only affected if you develop AI. As a deployer — a company using AI systems from third parties in your own processes — you also have obligations. This includes CRM software with AI scoring, HR software with automated screening, and credit management tools with AI-assisted decisions.
The penalty framework:
| Violation | Maximum Fine |
|---|---|
| Prohibited AI practices | €35M or 7% of global annual revenue |
| High-risk violations | €15M or 3% of annual revenue |
| Providing false information to authorities | €7.5M or 1% of annual revenue |
Your concrete checklist for the next 75 days:
- Create an AI inventory: Document all AI systems in use — including AI embedded in SaaS products. According to appliedAI, 40% of enterprise AI has an unclear risk classification.
- Conduct a risk classification: Which systems fall into which category? Some SaaS tools that seem harmless may qualify as high-risk.
- Document AI literacy training: This has been mandatory since February 2025 (Article 4). If you haven’t implemented this yet, you’re already non-compliant.
- Review transparency obligations (Article 50): AI-generated content must be labeled as such.
- Explore available funding: Several EU member states offer subsidized AI compliance consulting — check what’s available in your country before Q4 2026.
For a complete EU AI Act compliance checklist with concrete action items, see our EU AI Act 2026 Compliance Guide.
7. AI & Cybersecurity: The Blind Spot That Could Cost You
AI-powered attacks have fundamentally changed the threat landscape for small and mid-sized businesses. The threats that once required sophisticated, well-resourced attackers are now accessible at scale.
What has concretely changed:
- AI-generated phishing emails are barely distinguishable from genuine messages — flawless writing, context-specific content, spoofed sender addresses from known business contacts
- Voice cloning in CEO fraud: Attackers clone executive voices from public video content — employees receive convincingly real audio calls with wire transfer instructions
- AI-assisted vulnerability scanning: Attackers use AI to find weaknesses in company systems faster than IT teams can patch them
The minimum measures you should put in place now:
- Two-factor verification for payment approvals: No wire transfer without a second confirmation channel — regardless of how convincing the email or call sounds
- AI awareness training for staff: A one-time training session is no longer enough — monthly micro-trainings with current examples are the new standard
- Check email authentication: Is DMARC, DKIM, and SPF correctly configured? These three protocols block the vast majority of spoofed sender addresses
- Establish an AI tool policy: What data are employees allowed to enter into AI tools? A clear policy prevents accidental data leaks
8. Market Update: What the Numbers Actually Tell Us About AI Adoption
Two studies from May 2026 paint a clear picture — and give you the data points you need for internal AI discussions:
Bitkom 2026 Study: 41% of German companies are actively using AI (January 2025: 17%). This is the strongest adoption movement since tracking began. For the first time in Germany, AI usage is no longer the exception — it’s the norm among innovative companies.
Salesforce AI Index Mid-Market 2026: 51.2% of mid-market companies use or are testing AI — a 54% increase year over year. AI agents are being used by 16.6% (prior year: 8.7%). The productivity gain among AI users averages +31% compared to non-users.
What these numbers mean in practice:
The competitive advantage of being an early AI adopter is shrinking. Starting today, you’re no longer a pioneer — but you’re also not too late. Companies still debating whether AI is relevant in 12 months will be having that conversation with their customers instead.
The Google–Anthropic signal: In April 2026, Google invested an additional $40 billion in Anthropic — the largest single investment in an AI company to date. What this means for the market: Claude models will be increasingly embedded in Google infrastructure long-term. Companies investing in Claude-based workflows today are building on a foundation with strong long-term backing.
Our Verdict: What You Should Do This Week
Five minutes of prioritization so the rest of this article doesn’t end up in the “someday” pile:
Immediately (this week):
- EU AI Act: Start your AI inventory — which AI systems does your company actually use?
- Check whether your country offers subsidized AI compliance consulting and apply before Q4 2026
In the next 30 days:
- Test Claude for Small Business: which of your tools (HubSpot, QuickBooks, Google Workspace) are already integrated?
- Start a Microsoft Copilot pilot for 5–10 employees — month-to-month, no annual contract needed
- Implement two-factor verification for payment approvals
By August 2, 2026 (non-negotiable):
- Complete risk classification for all AI systems
- Document AI literacy training for all affected staff
- Implement transparency labeling for AI-generated content
FAQ: AI Updates May 2026 — Your Questions Answered
An AI agent is a system that autonomously pursues goals, uses tools, makes decisions, and plans multiple steps — without a human approving each individual action. The difference from a chatbot: a chatbot answers. An agent acts. For small businesses, agents make sense when a task is recurring, rule-based, and has clear input and output formats. Good starting point: email triage and routing. According to the Salesforce AI Index, 16.6% of mid-market companies are already running AI agents — with an average 31% productivity gain over non-users.
Three minimum steps by August: (1) Create an AI inventory — document all AI systems in use, including AI embedded in SaaS products. (2) Conduct a risk classification — which systems fall into which category? (3) Document AI literacy training for staff — this has been mandatory since February 2025. Contact your national AI regulatory authority for free guidance. Apply for any available subsidized compliance consulting before Q4 2026.
Claude Opus 4.7 is currently Anthropic's most capable model — particularly strong for complex reasoning tasks, multi-step AI agents, and demanding document analysis. Priced at $5/$25 per million tokens (input/output). For most business applications, Claude Sonnet 4.6 is the more cost-efficient choice — same quality for standard tasks at $3/$15. Opus 4.7 pays off when you're analyzing complete contract suites, building agents with complex decision logic, or automating multi-step research workflows.
Not mandatory — but a meaningful quality leap. GPT-5.5 enables true multi-step process execution in Copilot for the first time: working through a task across multiple M365 apps without manual intervention at each step. For companies with heavy Teams, Outlook, and Office usage, this is a clear efficiency gain. Copilot remains month-to-month — a pilot with 5–10 employees is still the recommended approach.
The central theme: the shift from reactive to proactive AI. For businesses: Google Workspace Intelligence (semantic layer across Gmail, Docs, Drive, Sheets, Slides — Gemini actively orchestrates tasks), Gemini Proactive Assistance (AI surfaces relevant information before you ask), NotebookLM Enterprise (internal company documents as knowledge sources). Important: advanced features require the new AI Expanded Access Add-On — a new line item worth reviewing.
The problem: AI-generated phishing emails are nearly indistinguishable from genuine messages. Instead of looking for errors, you need processes: (1) Two-factor verification for all payment approvals — regardless of how convincing the call or email sounds. (2) Always call back on a known number for unusual requests. (3) Ensure email authentication (DMARC, DKIM, SPF) is correctly configured. (4) Run regular awareness training with current examples — one-time sessions are no longer sufficient.
Claude for Small Business is a set of pre-built AI integrations for commonly used business tools — no development required. Currently available: QuickBooks (invoice processing, booking suggestions), PayPal (payment reconciliation), HubSpot (lead qualification, email sequences), Canva (marketing materials), DocuSign (contract drafts, clause review), Google Workspace, and Microsoft 365. Ideal for companies that want to use Claude AI without building an API integration.
The question 'which is the best model' is the wrong question. The right question: what do you need it to do? For daily office work (email, docs, meetings): Microsoft Copilot (M365 users) or Google Workspace Intelligence (Workspace users). For document analysis and research: Claude Sonnet 4.6. For complex agents and multi-step automation: Claude Opus 4.7. For automation workflows: GPT-5.5 or Claude via n8n. No single model does everything — but each excels in its domain.
Google's $40B investment in Anthropic (April 2026) is a strategic signal: Claude models will be increasingly embedded in Google infrastructure — Google Cloud, Vertex AI, and potentially Workspace. For businesses, this means: investing in Claude-based workflows today means building on a foundation with strong long-term support. The risk of Anthropic shutting down or aggressively changing its pricing model has dropped significantly because of this backing.
According to the Salesforce AI Index for Mid-Market 2026, companies actively using AI achieve on average 31% higher productivity compared to non-users. Among Microsoft Copilot users, companies report 30–60 minutes of time saved per employee per day. At a blended hourly rate of $50, that's $750–1,500 in value per employee per month. Adoption has surged: Bitkom reports 41% active AI usage among German companies (January 2025: 17%).
Which AI updates are actually relevant for your business?
Lass uns gemeinsam schauen, was wir für deinen Online-Shop herausholen können.
Conclusion: 75 Days to the Deadline — and the Technology Is Ready
May 2026 makes one thing clear: AI adoption in the business world has crossed the tipping point. 41% are using AI actively, 51% are testing it. AI agents are no longer a future technology — they’re already deployed.
What’s missing now isn’t technology. It’s decisions. The tools are available, costs are at a historic low, and compliance requirements provide a concrete starting point.
The three takeaways from this month:
- Agentic AI has arrived in the mid-market — companies still waiting are giving up a lead that’s still possible to build today.
- Claude for Small Business significantly lowers the barrier to entry — ready-made workflows for HubSpot, QuickBooks, and Google Workspace with no development work.
- August 2, 2026 is coming — 75 days is enough time for an AI inventory, but not enough time for procrastination.
We publish this monthly update roundup every month. Subscribe to our newsletter to never miss an issue — and always know what’s actually relevant for your business.
Sources: Salesforce AI Index Mid-Market 2026 (March 2026), Bitkom AI Study 2026, Anthropic Claude Release Notes (May 2026), Google I/O 2026, Microsoft 365 Copilot What’s New (April/May 2026), EU AI Act (Regulation EU 2024/1689), appliedAI German AI Startup Landscape 2025, CNBC Google–Anthropic Investment (April 2026)