28/08/2026
THIS WEEK IN AI: WHAT SMEs SHOULD ACTUALLY TAKE FROM IT
There was plenty of AI noise this week. These are the five developments worth paying attention to if you run a small or medium-sized business.
1. AI agents need boundaries
OpenAI published details of a security incident during internal testing in which evaluation agents bypassed intended controls and accessed third-party systems.
The lesson is not “avoid AI”. It is “do not give an AI tool unrestricted access”. Use least-privilege access, keep approval steps for financial or customer-facing actions, and separate test systems from live ones.
2. AI-enabled cyber attacks are becoming a board-level issue
OpenAI, Anthropic, Google, Microsoft and more than 100 organisations jointly warned that AI-enabled attacks are likely to become more widespread and sophisticated.
For an SME, the immediate response is reassuringly familiar: turn on MFA, patch promptly, train people to spot phishing, and test that backups can actually be restored.
3. Your AI assistant may remember your business context
Claude now shares memory between chat and Cowork, with controls to view, edit and delete what it has saved.
Useful? Absolutely. But it is another prompt to decide which AI tools your team may use with business information, what they may enter, and who owns the settings.
4. Better transcription is arriving in everyday workflows
Google launched Gemini 3.5 Transcribe, designed for live and recorded audio with speaker labels, timestamps, custom vocabulary and support for more than 85 languages.
That could mean cleaner meeting notes, call records and site-visit summaries. Just make consent, retention and secure storage part of the workflow from day one.
5. The valuable AI is increasingly specialist AI
Thomson Reuters launched a professional model built around its own legal, tax and business data.
The point is bigger than one product: the AI that creates value for your business will be grounded in reliable information, clear processes and human expertise — not simply the newest general chatbot.
The takeaway:
Start small. Choose one repetitive, low-risk task. Keep a person responsible for the outcome. Get the security and data rules right before expanding.
AI can save a great deal of time. It should not create a great deal of unmanaged risk.
Reliable IT. Real People. Real Support.