In partnership with
Good morning, {{first_name | AI enthusiast}}.
Frontier AI just did something security researchers hadn’t managed in over a decade: it found two critical holes in PHP, the code running roughly 70% of the web. OpenAI, Wiz, Cloudflare, and PHP’s own maintainers spent two months quietly patching the bugs before anyone could exploit them.
It’s a genuine win for AI-assisted defense, but it also means the same models can hand attackers a shortcut to exploits that used to take years to engineer. So which side of that race wins out when every hacker has access to the same tools?
Today in AI Brief:
AI uncovers a web-breaking PHP vulnerability
Microsoft’s AI outdiagnoses doctors on hard cases
Anthropic’s Claude Duet works while you’re away
Leave Granola and get up to 12 months free of Wispr Flow Notetaker + Dictation
If you have paid time left on an individual Granola plan, we'll match it with a Wispr Flow subscription that includes Notetaker and dictation, and add bonus time, up to 12 months total. Sign in or create a Wispr account and submit proof of your plan to check eligibility.
AI Catches a Flaw That Could’ve Broken the Internet
In Brief: A pair of frontier AI models just helped expose two critical flaws in PHP, the programming language that powers roughly 70% of the world’s websites — a find OpenAI called "one of the most impactful security issues in recent memory." Security firm Calif used OpenAI’s GPT-5.6 Sol to independently uncover the bugs, which Wiz separately confirmed using the same model alongside Google’s Gemini 4 Argon.
The Details:
The decade-old bugs let attackers hijack servers with no login required, prompting PHP’s maintainers, OpenAI, Wiz, and Cloudflare to quietly prepare protections for two months before disclosing anything publicly.
Five major U.S. banks, including JPMorgan Chase and Wells Fargo, separately started restricting third-party AI agents after detecting automated systems operating inside customer accounts.
OpenAI says its models found the PHP flaws “before attackers did,” opening what it calls a Defender’s Window — but warns the same capability can just as easily help attackers build exploits.
Take Away:
Frontier AI just proved it can find software flaws faster than any human researcher, and that cuts both ways — a decade-old bug nobody caught is now fair game for defenders and attackers alike. Expect this AI-versus-AI security race to define the rest of 2026.
Microsoft’s AI Beat Doctors on the Hardest Medical Cases
In Brief: Microsoft’s new diagnostic system, MAI-DxO, correctly diagnosed complex medical cases in a simulation more than twice as often as generalist physicians, according to a study published in Science this week. Microsoft AI CEO Mustafa Suleyman called the result a step toward "medical superintelligence."
The Details:
On hard cases from the New England Journal of Medicine, MAI-DxO scored 83.2% accuracy, versus 36.1% for primary care doctors and general internists working alone with reference materials.
MAI-DxO works sequentially, simulating a panel of physicians that question a patient, order tests, and debate a diagnosis over several rounds while weighing cost.
Most of the gain traces back to the underlying model: OpenAI’s GPT-5 alone scored 78.9% on its own, with Microsoft’s orchestration adding only about four more points.
Take Away:
Microsoft is now testing MAI-DxO with real patients at Beth Israel Deaconess Medical Center in Boston, and that result, not the simulation, will determine whether a "panel of AI doctors" becomes a real clinical tool. Google and OpenAI are chasing the same claim, making hospital diagnostics one of the most competitive fronts in AI right now.
SPONSORED BY LOVABLE
Nobody's coming to approve your idea
People assume that real businesses get built by other people with more resources. It's how great ideas get stuck in notes apps and conversations.
But you don't need six months and a developer's salary to validate your concept. You just need a weekend. Describe what you want to exist and Lovable builds it: a real site, a real product, ready to accept real payments from real customers. Not a mockup to show people. An actual thing they can use.
You don't need anyone's sign-off to get started. Take $5 off your first month.
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
Anthropic’s Claude Duet Works While You’re Away
In Brief: Anthropic opened a waitlist for Claude Duet, a standalone app that lets Claude keep working in the background on its own virtual computer, even when a user’s phone or laptop is off. The research preview runs on Opus 5.5 and is rolling out to U.S. Pro and Max subscribers.
The Details:
Duet checks connected apps every few hours, sends a morning summary, and reaches out when it needs a decision, carrying over memory and connectors from a user’s existing Claude account.
The agent still needs sign-off before sending messages, making payments, signing documents, or deleting files, and it blocks trading securities or transferring money outright.
Anthropic poached ex-Meta product lead Karan Ahuja to build Duet, and rival Meta’s Muse had already topped 14 million installs as of this week.
Take Away:
Every major lab now has a bet on an "always-on" personal agent, and Anthropic is leaning on Opus 5.5 — widely seen as a top-tier model — to differentiate Duet from Meta’s Muse and OpenAI’s Dots. Whether proactive agents actually save people time, rather than just adding another inbox to manage, is still the open question.
Everything else in AI
OpenAI nears a $30 billion funding round at a $1.25 trillion valuation, led by UAE investors, as the company edges closer to a potential IPO.
Canva launched Craft, its first in-house image generation model, which plans each image’s layout before rendering and supports 55 languages.
Waymo plans to grow its robotaxi fleet from about 4,000 vehicles to 20,000 by the end of 2027, backed by a fresh $5 billion loan.
Getty Images filed for Chapter 11 bankruptcy to go private, after years of pressure from AI-generated imagery and a collapsed Shutterstock merger.