Research Period: 14 July 2026 to 21 July 2026
1. Google renames NotebookLM to Gemini Notebook and adds native code execution
What changed
Google has renamed NotebookLM to Gemini Notebook, and your existing notebooks carry over automatically with nothing for you to do. The bigger news sits underneath the rebrand: every notebook now has its own secure cloud computer that can run code on the documents you upload.
Why it matters
Until now, a notebook could read your files and answer questions about them. Now it can actually calculate. Upload a spreadsheet of monthly sales figures and it can work out totals, spot trends, and build comparisons, grounded in your real data rather than guessing. Think of it as having a numerate assistant who has already read every document you handed over. Crucially, this rolls out to Google Workspace business accounts, which breaks Google’s usual pattern of leaving business customers waiting. If you run your business on Workspace, you are included from the start. It is available across the UK and Europe with no regional restrictions.
How to use it
- Upload a CSV of monthly figures and ask it to calculate quarterly totals and growth rates.
- Feed it several research documents and have it summarise findings with the numbers checked.
- Analyse a customer feedback spreadsheet to count themes and rank the most common complaints.
- Compare two years of expense data by asking direct questions rather than building formulas.
- Turn a messy exported report into a clean summary with the maths done for you.
Source: Google Blog | TechCrunch
2. OpenAI discloses GPT-Red, an automated red-teaming system that outperforms human hackers
What changed
OpenAI has revealed GPT-Red, an internal model built to attack its own AI systems and find weaknesses before real attackers do. In testing, GPT-Red succeeded on roughly nine out of ten prompt injection attempts where human specialists managed just over one in ten. You will never use GPT-Red directly; it stays inside OpenAI. What reaches you is the result: GPT-5.6 Sol was trained against GPT-Red’s attacks, so Sol is now far more resistant to being hijacked.
Why it matters
Prompt injection is where an attacker hides instructions inside a document, email, or web page that your AI reads, hoping to trick it into acting against you. Following last month’s agentjacking story, this is the defensive side of the same problem. The benefit here is passive: no setup, no extra cost, and it is already working for everyone using Sol. It matters most if you are building agentic workflows where AI touches your email, CRM, or files, because those are exactly the places a hidden instruction could do damage.
How to use it
- Prefer GPT-5.6 Sol for any workflow where AI reads untrusted documents or external emails.
- Stay alert to prompt injection when connecting AI agents to your business systems.
- Explain to your team why the model choice matters for security, not just capability.
- Review which of your AI tools process content from outside sources you do not control.
- Treat improved resistance as helpful, not absolute, and keep sensible human checks in place.
Source: OpenAI Blog | MarkTechPost
3. AI is now the intermediary between your business and potential customers
What changed
A Harvard Business Review article argues that AI tools have become the first place many potential customers go to research suppliers, before your website and before a Google search. Per HBR, if someone asks Claude, ChatGPT, or Perplexity about your business and gets an inaccurate or incomplete answer, that is now a customer discovery problem rather than a technical curiosity.
Why it matters
This is the next chapter of the SEO discipline most guild members already know, only the target has shifted from Google’s rankings to what AI tools say about you when asked. The good news is the fix costs essentially nothing and you already have every tool you need. You simply audit how AI describes your business today by asking it directly, then correct the record on your own pages so the AI has accurate material to draw from.
How to use it
- Ask Claude, ChatGPT, and Gemini “Tell me about [Your Business] and what they do” and read carefully.
- Try Perplexity with “What do customers say about [Your Business]?” to see what surfaces.
- Note every gap, outdated fact, or inaccuracy in what the AI tells you.
- Update your About page, services descriptions, and FAQ with specific, accurate language.
- Repeat the audit quarterly, since what AI tools say about you will keep shifting.
Source: Harvard Business Review
Additional Noteworthy
Anthropic launches Claude for Teachers: Eligible US K-12 teachers get premium Claude access including Cowork and Code. Members in US education or L&D roles may qualify; UK and Europe members should watch for international expansion. | Source
Microsoft Scout, an always-on M365 AI agent: Announced at Microsoft Build in June, Scout watches your calendar and email to prepare meeting briefs, block time, and flag deadlines. Currently private preview only, needing Frontier programme enrolment and a GitHub Copilot licence. | Source
GPT-5.6 API pricing published: Sol costs $5/$30 per million tokens in/out, Terra $2.50/$15 (half of GPT-5.5), and Luna $1/$6. GPT-5.5 stays available in the API. We covered the capabilities last week; this is the pricing for anyone building on the API. | Source