BC sues OpenAI over school shooting, OpenAI solves 100+ math problems, and California's new data center laws

A Canadian province just walked into a San Francisco courthouse and pointed the finger directly at OpenAI. And separately, that same company's AI just quietly solved over 100 problems that have stumped mathematicians for years. It's one of those days where the legal and scientific storylines could not be more different - but both land hard.

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

British Columbia has filed a lawsuit against OpenAI and CEO Sam Altman in a San Francisco federal court, alleging that the company could have prevented a deadly mass shooting at a school in Tumbler Ridge - if it had warned police that the shooter used ChatGPT to plan the attack.

The lawsuit argues that OpenAI had access to information that indicated violent intent and failed to alert law enforcement. This is a significant legal theory - essentially arguing that AI companies have a duty to report when their tools are being used to plan crimes, similar to how some online platforms have obligations to report child exploitation material.

This case joins a growing wave of legal battles over what AI companies are responsible for when their products are misused. OpenAI has consistently argued it is not liable for how users apply its tools - a position courts have not fully tested yet. A ruling against OpenAI here could set a precedent that forces AI companies to actively monitor and report suspicious conversations, which raises its own enormous data privacy concerns.

Read the full story →

This is the most direct legal challenge yet to the idea that AI platforms bear zero responsibility for harms enabled by their tools - and its outcome could reshape how every AI company designs its safety systems.

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Download the Free Report

OpenAI has revealed that its AI has resolved more than 100 open mathematical problems - and in response, the company is forming a dedicated math advisory group to oversee and guide this research. The advisory group won't have authority to slow down or redirect OpenAI's ongoing math work, but it signals just how seriously the company is taking this new frontier.

This comes weeks after OpenAI separately claimed its AI agents solved the Navier-Stokes problem on September 8th - one of the seven Millennium Prize Problems, each worth $1 million to the solver. The Guardian editorial board has raised questions about how independent and genuinely useful the AI's mathematical work actually is, noting that when a human solves such a problem they collect both prize money and peer validation - neither of which has happened here.

For context on why this is significant: reasoning models have been steadily improving at formal mathematics, but resolving this many open problems in a concentrated period is a different category of result. If the math holds up under scrutiny from the broader academic community, it would represent a genuine acceleration in what AI systems can independently contribute to human knowledge - not just assistance, but discovery.

Read the full story →

The math results need independent peer verification before they become historic - but even the attempt at this scale is a signal that AI's role in pure research is shifting from assistant to potential co-author.

California Gov. Gavin Newsom signed a package of seven new bills this week targeting AI data center energy and water consumption. The core change: data centers can no longer pass the cost of grid and water infrastructure upgrades onto everyday utility customers. Tech companies will now foot those bills directly.

The package also requires California's Public Utilities Commission to create a new rate classification specifically for data centers - separating their massive energy appetite from residential and standard commercial customers. Other bills in the package address water usage disclosures and efficiency standards, as AI training and inference workloads put increasing strain on local water systems for cooling.

This is part of a broader California pattern - the state has been aggressive this year on AI infrastructure accountability. If you followed our coverage of California's AI kill switch push, this fits the same pattern of the state moving faster than federal regulators. Expect other states to watch California's rate classification approach closely.

Read the full story →

For AI companies expanding data center capacity, California just made the true cost of compute significantly more visible - and non-negotiable.

Researchers have developed a large language model called Apollo specifically designed to fill in the gaps in ancient Greek papyrus fragments. These are physical documents - often tattered, burned, or water-damaged - that contain records of everyday life, literature, and administrative details from the ancient world. Scholars have spent careers trying to reconstruct partial texts, and now an AI is stepping in as a reading partner.

Apollo works by learning the patterns of ancient Greek writing, then predicting what letters or words are likely to fill lacunae - the technical term for gaps in the surviving text. The model was trained on existing transcribed papyri and can generate statistically probable completions based on context, script style, and linguistic patterns of the period.

The practical upside is speed. A papyrologist might spend weeks analyzing a single fragment's gaps. Apollo can generate candidate completions in seconds, giving researchers a starting point to evaluate rather than a blank space to stare at. It's a compelling example of language models being applied to genuinely difficult, narrow domains rather than general-purpose chat. And if you're building a tool for a niche expert audience, this is a useful model - pun intended - for what focused AI products can look like. Speaking of building tools fast: 60sec.site lets you build an AI-powered website in under a minute, no coding needed.

Read the full story →

Apollo is a reminder that some of AI's most interesting applications are not in consumer apps - they're in the hands of a few hundred specialists who suddenly have a supercharged research partner.

The answer is The Navier-Stokes Problem! On September 8th, OpenAI claimed its AI agents solved this famous fluid dynamics problem - one of the seven Millennium Prize Problems worth $1 million each. The mathematical community is still evaluating the result, but it is the most high-profile claim in OpenAI's recent mathematical push.

The BC lawsuit raises a genuinely hard question: should AI companies be required to report when users appear to be planning harm? Or does that create a surveillance system worse than the problem it solves? Hit reply and tell me where you land - I read every single response.

That's all for today. More AI news tomorrow - find the full archive any time at dailyinference.com.

Know someone who would like this newsletter? Share your link. One referral gets you the AI Tools Starter Kit, and five referrals get you the AI Insider Briefing.