Google found 10 bugs no other AI could find
Two stories this week that change how builders think about AI and the law.
Two things happened this week that most people glossed over.
One changes how companies protect their code.
The other changes how the entire AI industry thinks about copyright.
Both matter to you. Here is what you need to know.
Google just made AI security scanning cheap enough to run constantly
Security has always been expensive. Not just in tools but in time.
Finding vulnerabilities in a large codebase used to mean paying for expensive frontier models to do slow, costly scans. Most teams ran them once a quarter if they were being diligent. Many never ran them at all.
Google just changed the math.
Gemini 3.5 Flash Cyber is a cheaper, faster security model built specifically for repeated vulnerability scanning and automated patching. It is launching alongside Gemini 3.6 Flash as part of Google’s latest model lineup.
The numbers are worth paying attention to.
In V8 testing, the model found 55 confirmed security issues. Ten of those were not discovered by any other model. Not frontier models. Not expensive alternatives. Nothing else caught them.
The key word here is repeated. Because Gemini 3.5 Flash Cyber is built to be called multiple times quickly and cheaply, CodeMender can explore more code paths per scan than larger systems can afford to. Speed and cost are the features, not afterthoughts.
Google’s direct target here is Anthropic’s Mythos 5, which has positioned itself as the premium option for cybersecurity work. Gemini 3.5 Flash Cyber is not claiming to be better. It is claiming to be good enough, faster, and significantly cheaper.
For founders and builders this means one thing.
Security scanning that used to require enterprise budgets is becoming accessible to small teams. If you are building anything that handles sensitive data, this is worth watching closely.
Gemini 3.5 Flash Cyber will initially be available to governments and trusted partners through CodeMender. Broader access is expected to follow
Anthropic just settled the largest copyright case in AI history
This one is bigger than most people realize.
Anthropic has reached a $1.5 billion copyright settlement that has now been approved by a judge. The case was brought by authors and publishers over millions of books used to train Claude without permission.
The settlement pays $3,000 per work across an estimated 500,000 works. Authors and publishers receive the money directly. It is believed to be the largest US copyright settlement ever.
But here is what makes this complicated.
The judge ruled that training AI on copyrighted text can qualify as fair use. That part of the ruling is significant and potentially favorable for the entire industry.
However the same judge also found that Anthropic illegally obtained books from pirate sites including Library Genesis and Pirate Library Mirror. That distinction matters. The fair use argument held for legitimately sourced content. The piracy route did not.
The settlement closes this specific case without creating binding nationwide precedent. That is the most important sentence in this story.
Lawsuits against Google, Meta, Midjourney, and OpenAI are still moving through the courts. Every one of those cases will be influenced by how judges interpret what happened here. But none of them are automatically bound by this ruling.
What this means practically for AI builders is that the legal foundation of training data is still being written in real time. The industry does not have a clean answer yet. What it has is one very expensive data point that suggests the courts will draw a distinction between how content was sourced, not just whether it was used.
That nuance will matter for every company building on AI trained data for the next several years.
Featured Tools
The tools serious builders are using right now.
FocuSee - A product demo maker that automates video editing with AI-powered enhancements. Turn raw screen recordings into polished, share-ready demos with smart zooms, 3D motion effects, and clean layouts.
Setapp - Launch your Mac app in front of paying users while Setapp handles billing, licensing, taxes, and distribution.
Atoms.dev - Multi-agent AI that helps you research, build, and grow products faster.
Shift AI - AI built directly into your browser workflow so you can summarize, write, and research without switching tabs.
❤️ Founders & AI Builders — Get Featured
If you’ve built an AI tool and want to reach 60,000+ AI-first readers.you can now sponsor an upcoming AI Cheatcode issue.
👋 Reach out here to collaborate.
📩 Quick Favor
If this email landed in your Promotions tab,
drag it into Primary so you don’t miss the next drop of AI news, tools, and automation blueprints.