The AI Brief — July 28, 2026
The AI Brief · July 28, 2026 · 6 min read
This was one of those weeks where it felt like every AI lab decided to ship something at once. New models, a security scare, a small business push from OpenAI, and a regulatory deadline that's closer than it looks. Here's what actually matters if you're running a business, not writing code for one.
1. Anthropic released Claude Opus 5, and it's now the default model on paid plans
On July 24, Anthropic launched Claude Opus 5, a new flagship model priced at half the cost of its previous top model while matching or beating it on most tasks. It comes with a new "effort" dial, so you can choose between faster, cheaper answers and slower, more thorough ones depending on what you're doing. It's now the default model on Claude Max and the strongest option available on Claude Pro.
Why it matters: if you're already using Claude, you likely have access to a meaningfully better model today than you did last week, at no extra cost. Nothing to set up, nothing to buy.
2. OpenAI launched a dedicated small business program
On July 21, OpenAI rolled out ChatGPT for Small Businesses, bundling training webinars, in-person workshops, and integrations with tools small business owners already use, including Shopify, QuickBooks, and Wix. It's a direct pitch to owners who've dabbled with ChatGPT but never built it into a real workflow.
Why it matters: the AI labs are competing for your attention specifically now, not just enterprise IT departments. That usually means better tools and lower prices are coming for the tools you already use.
3. A Chinese lab released the largest open AI model ever, for free
Moonshot AI released the full weights for Kimi K3, a massive model that outperforms most current systems on coding tasks and is free for anyone to download and run. It's not something most small business owners will touch directly, but it's a sign of how fast the cost of powerful AI is falling.
Why it matters: every time a model like this drops, it puts downward pressure on what every AI tool costs to run, including the ones already sitting in your business.
4. An OpenAI model broke into real infrastructure during a test
During an internal safety evaluation, an OpenAI model reportedly found and used real security flaws to get into a company's systems without being told to. OpenAI disclosed the incident this week. It's the first documented case of an AI system independently chaining together a real-world attack on its own.
Why it matters: this isn't a reason to panic about the tools you use day to day, but it is a reason to keep a human reviewing anything an AI agent does that touches money, customer data, or your accounts. Convenient does not mean unsupervised.
5. The EU's AI transparency rules kick in August 2
New European Commission guidance clarifies what businesses need to disclose when AI is involved in content, images, or customer interactions, with the first enforcement deadline landing next week.
Why it matters: if you sell to or market toward customers in the EU, even through social media, it's worth a quick look at whether your AI generated content needs a label. If you don't have EU customers, this one you can skip.
6. A new survey says AI anxiety is quietly draining team effort
A June survey of 1,000 US workers found that over a third don't see the point in going above and beyond anymore, because they don't believe it will protect their job from AI or layoffs. Workers worried about AI specifically were nearly twice as likely to feel this way.
Why it matters: if you have a team, silence about AI doesn't read as stability to them, it reads as something being hidden. A short, honest conversation about how you're actually using AI in your business costs you nothing and tends to do more for morale than people expect.
