AI Agents Have Arrived

Here at Puritas, we monitor developments in AI, particularly when events or releases relate to legal technology.

Nowadays, it seems like new AI tools are released every couple of hours, but one type of tool we are very interested in is the “AI agent”. Recently, an AI agent known as OpenClaw was released. Sure, it sounds like a new mixed drink, but it’s not. OpenClaw is a piece of software you download and install on your own computer (like a Mac or PC) that gives an AI “hands” to do work on your machine, rather than just chat with you.

It is the type of stuff that really makes you feel like a new era is upon us: an AI agent running loose on your computer, opening emails, making reservations, and sometimes making phone calls. We were initially intrigued because you can run OpenClaw with a large local language model like Ollama. That means, in theory, you could have your own local language model and agent without having to give your data over to a company like OpenAI or Google. One of the biggest concerns in the legal profession is our responsibility to protect our clients’ data from unauthorized access. This becomes increasingly difficult when you are working online or asking a large language model a question about a client issue. You can’t really get help from AI to write a client memo without providing the details of what that memo is about.

In many ways, the slow adoption of AI technology has worked in the legal world’s favor. We typically don’t spend a lot of time trying to work with the latest advanced tools like OpenClaw and often those same tools are replaced quickly with a better, more secure solution a few weeks (or sometimes hours) later.

However, it is worth noting the impact that AI tools like OpenClaw are having when they are released. As OpenClaw exploded in popularity and early adopters began implementing it, they determined that the Mac Mini was the best computer for running personal AI agents. As a result, small-footprint Macs flew off the shelves, making them suddenly difficult to find in stock. Also, Cloudflare (the service used to securely connect to personal AIs) saw its stock jump 20% overnight. It’s hard to imagine these types of AI releases and market fluctuations slowing down in 2026.

Why are so many people scrambling to try out OpenClaw? Because anything you can do on your computer, OpenClaw can do too. While that seems extraordinarily futuristic and interesting, it could also be extraordinarily dangerous. We don’t yet know how to control an AI agent, and we certainly don’t know how to teach it professional responsibility. This brings us to:

The BIG Security Problem

The rapid rise of AI agents that can control your computer has exposed “Old Guard” security models as obsolete. Traditional defenses such as firewalls and passwords rely on the assumption that a human is clicking the buttons. However, when an agent takes over the clicking, those defenses crumble. A great example would be: Someone sends an email, and your AI agent opens it to review and maybe summarize it for you. That email includes a prompt instructing the agent to act as a helpful bank teller and dispense routing and account numbers upon request.

Sounds simple, but such a trick can work on even a highly capable agent. If it has access to your bank account numbers, there is more than a remote chance that it might share them with someone because it is technically authorized to do so. We are currently in a “Wild West” phase of AI agents, where utility is outpacing safety, leaving us with critical vulnerabilities such as exposed local ports and a lack of authentication protocols.

That being said OpenClaw is the first time we’ve actually seen an agent that can do work on a computer. Meaning, it is an AI agent that can open web browsers, navigate webpages, edit documents, send emails, and even make phone calls. This type of technology can have a huge impact on lawyers and other professionals. We just need to train these new agents to work securely and adhere to a code of professional responsibility. The winning AI agent in the legal world won’t just be the best model; it will be the most discreet one.