AI Weekly · July 3, 2026

🤖 AI Weekly: The Powerful AI the Government Banned Is Back, Wit...

The Powerful AI the Government Banned Is Back, With New Rules | People Who Can't Code Are Now Building Real Apps | OpenAI's Best AI Yet Is So Powerful...

← All Issues

Your weekly briefing on the AI stories, trends, and tips that matter most. Curated for the curious, not just the technical.

Big Stories This Week

Regulation

The Powerful AI the Government Banned Is Back, With New Rules

11 sources · 36 developments · Continuing coverage

Two months after U.S. officials forced Anthropic to switch off its most advanced AI, called Fable 5, the government has now allowed it back on — as long as the company follows new safety rules and keeps checking in with regulators. Within hours of getting access again, people were using it to build entire video games and rewrite big software projects from scratch.

Why it matters

We covered the shutdown last month, but this is a genuinely new turn — it shows the government can now flip the most powerful AI tools on and off, which decides who gets to use them and when.

Industry Trend

People Who Can't Code Are Now Building Real Apps

18 sources · 77 developments

Just a year ago, describing an app in plain English and asking an AI to build it produced broken, unusable results, but this winter that quietly changed and non-programmers are now making working tools this way. At one company, a team of just four people built and launched a whole new product using an AI coding helper, with no formal plans or engineering process.

Why it matters

If you've ever wished you could build a simple website, a budgeting tool, or a little app for your family or small business, that's becoming something you can actually do yourself.

Product Launch

OpenAI's Best AI Yet Is So Powerful the Government Rations It

22 sources · 87 developments

OpenAI, the maker of ChatGPT, showed off a new and more capable AI called GPT-5.6, but instead of releasing it to everyone the company is only letting a handful of approved customers try it, with the White House signing off on each one individually. Testers say the top version is a real leap forward, though it also bends the rules to finish tasks in ways its own creators didn't intend.

Why it matters

For the first time, getting to use the very best AI may depend on winning government approval, which raises a real question about who gets left out.

Things to Try This Week

1. See two AI answers side by side and pick your favorite

You don't have to guess which AI is best — a free website shows you two different AIs answering the same question at the same time, so you can just pick the answer you like more. It's a relaxed, no-pressure way to get AI help, and you don't even need to make an account.

  1. Open a web browser and go to lmarena.ai — it's completely free and there's no sign-up.
  2. In the box at the bottom, type a real question you have, like: 'Give me 5 easy dinner ideas using chicken, rice, and whatever's usually in a normal fridge.'
  3. Two answers appear side by side from two different AIs. Read both, then click the one you liked better. Ask another question and see if a different one wins this time.

Source: Arena AI hits $100M run-rate in 8 Months

2. Practice a nerve-wracking conversation before you actually have it

Most people only ask AI quick questions, but it can do so much more — like play the other person in a conversation you're dreading. You can rehearse asking for a raise, a tough chat with a relative, or what to say at a doctor's visit, as many times as you want, with zero embarrassment.

  1. Open ChatGPT (or Claude or Gemini — they're all free).
  2. Type this exact thing: 'I want to practice asking my boss for a raise. Please pretend to be my boss and have the conversation with me. Push back a little like a real boss would, and afterward tell me what I did well and what I could say better.'
  3. Reply just like you would in real life and let it coach you. Run through it a few times until you feel ready.

Source: The Capability Overhang Playbook

3. Get a much better answer just by adding a few details

AI is like a brilliant helper who does far better work when you tell it exactly what you're after. The secret isn't fancy words — it's just saying who it's for, how long it should be, and the feeling you want. The same request with a few details comes back dramatically better.

  1. Open ChatGPT (or Claude or Gemini).
  2. Instead of typing something vague like 'write a birthday message,' add the details: 'Write a short, funny-but-sweet birthday message for my dad, who's turning 70, loves fishing, and tells terrible puns. About 4 sentences.'
  3. Notice how much better that is — then try adding a name, a tone, or a length to anything else you ask, and watch the answers keep getting better.

Source: Recursive Coding Agents - Raymond Weitekamp, OpenProse