Police Cameras May Be Watching More Than You Think — And the Code Proves It
Today’s AI stories share a thread: the gap between what AI systems are said to do and what they actually do. A surveillance company’s hidden capabilities just got exposed. Google is quietly repositioning its AI as a classroom tool. And developers are quietly patching the security holes that quietly accumulate in the software holding all of this together.
Flock’s Police AI Does More Than the Company Let On
Flock Safety makes cameras you’ve probably driven past without noticing — the ones mounted near highway on-ramps and parking lots that read license plates for police departments. The company has always pitched that narrow capability publicly. But Wired obtained and analyzed Flock’s actual code and found the system does considerably more.
According to Wired, the AI can identify and track individuals using characteristics beyond license plates. Think physical appearance, movement patterns, and other signals that tie a person to a location across time. That’s a meaningful difference. Knowing which car was somewhere is one thing. Knowing who was somewhere — and building a picture of their movements — is something else entirely.
The concern here isn’t just philosophical. Automated surveillance systems like this work by correlating data points across many cameras and many moments, building profiles without any single officer deciding to follow someone. That process can happen invisibly, at scale, with no warrant and often no public awareness. When a company’s public description of its tools doesn’t match what the code actually does, the people being watched have no way to understand their own exposure.
Why this matters: If accurate, this suggests police departments may be deploying AI capabilities their communities never approved or even knew about. That’s a transparency problem with real consequences for civil liberties.
“AI system can identify and track people in ways beyond previously disclosed capabilities”
Google Is Going After Students With AI Study Features
Google reportedly added a set of study-focused tools to both Gemini — its AI assistant — and Google Search, according to TechCrunch. The features appear designed to help students learn actively rather than just retrieve answers.
The timing is deliberate. AI tutoring has become a real battleground, with companies like Khan Academy, Chegg, and various startups competing for the attention of students who increasingly turn to AI first when they’re stuck on a problem. Google already has Search embedded in most students’ daily habits. Adding AI features on top of that existing behavior is a lower-friction move than asking students to adopt a brand-new app.
For students, this could be genuinely useful if the tools prompt thinking rather than just hand over answers. The difference matters: an AI that asks “what do you think happens next in this equation?” builds understanding, while one that just solves the equation for you doesn’t. Whether Google’s implementation leans toward the former or the latter will determine whether this is a real learning tool or just a homework shortcut with better branding.
Why this matters: Google reaching students through tools they already use every day means AI-assisted learning is becoming a default, not a choice — and that shift is happening faster than most schools have prepared for.
“Google adds new study tools to Gemini and Search for students”
A Security Patch for OpenAI’s Python Library
OpenAI released version 3.3.1 of its Python library — the software that developers use to connect their apps and products to OpenAI’s AI models. This wasn’t a feature update. The release on GitHub describes it as a maintenance update focused on fixing security vulnerabilities in the library’s dependencies.
Dependencies are external code packages that a piece of software relies on to function — think of them as pre-built components borrowed from other developers. When one of those borrowed components has a security flaw, every app built on top of it inherits that risk. Patching them promptly is unglamorous but essential work.
For anyone who doesn’t write code, this story is still relevant. Thousands of apps and services are built on top of OpenAI’s library. A vulnerability in that foundation could affect products many people use without knowing OpenAI is powering them. Security patches like this one are how that risk gets quietly managed before it becomes a problem.
Why this matters: Most people interact with AI through products they didn’t build. The safety of those products depends on developers keeping this kind of foundational software current.
“Maintenance release focused on security vulnerabilities in dependencies”
Also Happening in AI
Anthropic pushed out two updates to its Python SDK in quick succession — version 0.124.0 added Files and Skills APIs for general use, and version 0.125.0 added support for managed agents, which are AI systems that can handle multi-step tasks with some autonomy. Both are available on GitHub. Meanwhile, LangChain — a popular framework developers use to build AI-powered applications — released updates to both its core library and its OpenAI integration package, with version 1.6.0 of each adding better error handling and improvements to how AI tools process instructions. And Hugging Face, the platform that hosts thousands of open AI models, quietly released a patch fixing image processing bugs in its widely used Transformers library.
What to Watch
The Flock story is part of a pattern worth tracking closely: AI capabilities expanding quietly inside existing products, outpacing the policies meant to govern them. Watch for more investigative reporting on surveillance tech this fall, as city councils in several states are reportedly revisiting contracts with camera companies. On the student AI front, back-to-school season will be the first real test of whether Google’s new features change how students approach homework — and how teachers respond when they do.