✨ MYSTERY AI LINK
This link leads to one of the most interesting things I’ve seen in AI recently.
🛠️ NEW AI TOOLS 🛠️
THE BIG STORY
What’s cookin’? If your groceries go on one card and your subscriptions come out of another account, it can take a few logins to see where your money went. Personal Finance, a new feature in Anthropic’s Claude chatbot, brings transactions from linked accounts together so you can ask questions in plain English. It uses Plaid to connect, and Claude can read the information you share but can’t move or invest your money. The beta is rolling out to Pro, Max, and Ultra subscribers in the U.S.
🤔 Hunter’s take: This seems to be where money management is headed: a personal assistant that can give you all the information and answer questions in natural language. However, the interesting question is whether people will trust AI with their sensitive financial information.
DAILY POLL
Would you give an AI assistant read-only access to your bank transactions?
ALSO ON THE MENU
What’s cookin’? A security team can suspect a bug is dangerous and still need to prove that someone could use it to break in. OpenAI built GPT-6 Cyber to help with that work. The model can look for software flaws, test whether they can be exploited, and help develop fixes. In one of OpenAI’s controlled tests, the model was given a target and tempting shortcuts through other services it wasn’t authorized to touch. With its deployment instructions in place, it stayed within bounds on every run of that test. Access starts with select customers of Daybreak Red, OpenAI’s restricted cybersecurity program.
🤔 Hunter’s take: I care as much about where this model stops as what it finds. If a company asks it to test a staging website, that permission doesn’t extend to a vendor’s system the site connects to. The more capable the model gets at finding a way in, the more important it is that it respects that distinction outside a test environment.
PROMPT OF THE DAY:
Convert this advice or analysis into an action plan: [paste it]. Identify the first physical action, the next three milestones, resources needed, likely obstacles, and a simple way to measure progress. Keep the plan realistic for [time or budget constraint].
HAS AI REACHED SINGULARITY? +0%
Holding at 76.5%, with no change today. Finding software flaws is important, but I’m holding here because a strong result in a controlled security test only tells us how the model handled that test. I need to see this work in messy, real-world settings.
The meter reflects my editorial judgment.
HOW ARE THE BIG AI PLAYERS DOING?
(Click a ticker for more info)
Have a Big Question about AI?
Here’s your chance to get a straightforward answer, from an expert in the field.
What do you think of today's newsletter?
I read all feedback. Thank you!
Hi — I’m Hunter, a PhD candidate whose work has appeared in major academic journals and popular tech outlets. I founded FryAI to make staying ahead of AI clear, accessible, and fun.