Free AI models, local laptops, and the guy who tried to hack a judge 🚀⚖️

Hey everyone,

Welcome back. This week on the podcast, we dug into two massive shifts happening right now: the bottom falling out of the AI pricing market, and the legal sector getting a very messy crash course in algorithmic vulnerabilities.

If you want the full breakdown, you can listen to this week's episode here: Welcome to the AI Price War.

Otherwise, here are the core takeaways you need to know.

The Bottom Falls Out on Pricing

The AI pricing war is officially a race to zero. OpenAI just handed out unlimited access to its GPT-5.6 models for free.

As I mentioned on the show, giving away a frontier model at zero cost completely flips the board for grassroots tech communities and non-profits. These organizations usually operate on shoestring budgets. Free access to top-tier reasoning engines removes the financial barrier to building community-scale automation. With competitors like Elon Musk's Grok severely undercutting the market, the days of developers being locked into a single expensive API are over.

Taking the Models Home

We are also watching a massive spike in developers taking their infrastructure offline. Platforms like Hugging Face have made it incredibly simple to download models directly to your own hardware.

Running a lightweight model on your laptop isn't just a fun weekend experiment; it’s a strategic privacy move. It allows you to build, test, and iterate without sending proprietary data to a cloud server or paying a meter.

Big Law Meets Big Data

On the enterprise side, professional services are getting serious infrastructure upgrades. Global law firm Kirkland & Ellis teamed up with Palantir to launch a custom AI platform for private equity fundraising.

They are essentially trying to consolidate fundraising documents, market data, and institutional knowledge across a network of over 1,000 lawyers into a single, queryable model. Grainne summed it up perfectly on the podcast: this is a huge leap for an industry that has historically kept AI at arm's length. But turning private equity data into a centralized AI model immediately raises high-stakes questions about client confidentiality and data siloing.

Hacking the Court

Finally, we have our first documented case of someone trying to execute a prompt injection attack against the U.S. legal system.

In a recent Connecticut case, a litigant attempted to manipulate the court's automated document scanners (source). He embedded hidden instructions in his legal filings using white text on a white background. A human couldn't see it, but the hidden text ordered any AI system reading the document to rule in his favor and treat a previous negative ruling as an error.

The judge spotted weird formatting anomalies on the printed pages, figured out what was happening, and revoked the guy's electronic filing privileges. It's a slightly hilarious but deeply alarming warning: if you are going to use AI to process important documents, you need aggressive governance over your inputs.

Thanks for reading. We’ll see you next week when we interview our first podcast guest of the season. :) Be sure to listen in!

Fleur & Grainné

AI Leader & Tech Strategist