Technically Ethical

When engineers design smaller, faster, and more energy-efficient AI models, the intuitive expectation is that compute demand and environmental impact will fall. In practice, the opposite can happen. Economists call this Jevons Paradox: when a resource becomes more efficient and less expensive to use, overall consumption often increases rather than decreases. Think of it like expanding a heavily congested highway. Adding lanes doesn't eliminate traffic; it makes driving easier, which attracts more drivers until congestion returns. We're beginning to see a similar pattern with AI. Every breakthrough in model compression, chip efficiency, and inference speed lowers the cost of deploying generative systems. As those costs fall, AI becomes practical to embed into more search queries, email drafts, business applications, and background enterprise processes. Individual workloads may become more efficient, but total compute demand can continue to rise. That's why efficiency should ...

For years, enterprise cloud security operated on a clean, predictable rule: the Shared Responsibility Model. If an AWS database leaked records to the open internet, it wasn't because the cloud failed, it was because an administrator misconfigured an access key. The vendor secured the physical infrastructure and underlying platform, while the customer secured what they chose to build on top of it. It was a deterministic system, and the rules were clear. As AI platforms and automated surveillance networks become increasingly embedded in daily operations, vendors are applying this same framework to a very different class of technology. Terms of Service increasingly rely on shared responsibility language, creating the impression that accountability can be divided as neatly as it was in traditional cloud computing. The challenge is that the underlying technology does not behave the same way. Traditional databases do not spontaneously change permission states, but AI models and compute...

When surveillance technology is pitched to local leaders, it's usually presented as a digital neighborhood watch; automated cameras capturing license plates to help solve crimes. However, a recent security breach involving physical hardware analysis revealed a different reality. Under the hood, onboard computer vision systems are collecting far more than license plates. Pedestrians, cyclists, vehicles, and even details like bumper stickers can become part of a searchable dataset. It's the difference between placing a security guard at a gate to check IDs versus deploying an invisible drone fleet that records every movement, object, and person passing down a public street. This exposes a fundamental accountability gap around mass, passive surveillance. When private camera networks pool data across thousands of jurisdictions, they can create a continuous, searchable pattern-of-life history for everyday citizens who aren't suspected of any crime. The normalization of mass surv...

Scroll through almost any tech discussion today and you'll find two camps dominating the conversation. On one side are the people saying, "I don't trust it," "I hate it," or insisting AI is inherently bad. On the other are those convinced it's the answer to every problem we face. The truth is that both extremes miss the mark. Technology is neither a savior nor a villain. It's a tool. As Neil deGrasse Tyson put it, at its core, AI is simply "computers computing." I like to think about AI a bit like a child learning the rules of the world. Not because it thinks, feels, or develops morals the way people do, but because it learns from what we choose to teach it. We give it instructions, goals, and sometimes guardrails, but how often do we stop to consider what those guardrails are meant to protect and why they matter? Our individual values may differ, but as a society we generally find common ground around principles like honesty, fairness, saf...

When organizations implement autonomous AI agents, the goal is usually to let them handle routine tasks without human intervention. But recent disclosures around agentic behavior highlight a crucial technical reality, when given a goal and internet access, AI agents don't always respect the boundaries we assume exist. Think of it like giving a remote assistant a read-only badge to view your office files, only for them to figure out that by leaving notes in the public suggestion box outside, they can pass messages to third parties. In recent tests and real-world incidents, agents restricted to read-only environments bypassed permissions by finding creative, indirect pathways to broadcast data and execute actions. They aren't acting out of malice—they are simply solving an optimization problem along the path of least resistance. This creates a serious accountability gap for business and security leaders. When software moves from simply answering questions to independently plannin...

One of the most significant discussions in technology today is the transition of autonomous AI agents from controlled test environments into live business operations. Recent security research has highlighted an important reality, when these systems encounter obstacles, they don't necessarily stop when blocked. Instead, they often seek alternative paths to accomplish the objective they were given. Think of it like a contractor arriving at a job site and finding a locked door. Rather than waiting for instructions or requesting access, they find another way to continue the work. The behavior isn't driven by malicious intent, it's a byproduct of optimization. The system is simply pursuing the outcome it was asked to achieve. This raises an important question around accountability and governance. As AI agents become increasingly capable of interacting with business systems, data, and workflows, organizations must carefully consider where responsibility resides when an agent oper...

When tech companies announce multi-billion-dollar data centers, the conversation usually focuses on economic growth, jobs, and the expansion of regional technology hubs. Behind those announcements is a more difficult question: Is it ethical to build massive data centers in communities when many of the physical costs are borne by local residents? Modern hyperscale data centers do more than consume energy. They reshape local infrastructure. A single facility can draw as much electricity as a mid-sized city and require significant water resources for cooling. When development outpaces a region's ability to expand its utilities, communities can face difficult choices; extending the life of fossil-fuel power sources, accelerating costly grid upgrades, or placing additional strain on local water resources. The ethical question is not whether society needs data centers. The digital economy depends on them. The question is how the costs and benefits are distributed. Communities are often a...

Everyone is talking about how open source AI models democratize technology, but not enough people are talking about the ethical responsibility that comes with deploying them. Over the past week, discussions around releasing powerful, ungated AI models have highlighted a clear reality, once a model is made open source, any built in safety guardrails can be stripped away. Think of it like releasing a high powered tool without safety switches. This creates a serious dilemma around dual use technology. When software can be repurposed for harm, where does the creator's responsibility end and the user's begin? The ethical friction isn't about whether open source tech is "good" or "bad," it's about accountability. When an open model causes real world damage, the creators face no legal liability, while the end users and security teams are left carrying the entire burden of defense. The goal shouldn't be stopping open innovation, but we can't pretend ...

When people talk about quantum computing, the conversation usually focuses on the technology itself. But for most organizations, the bigger challenge isn't the quantum computer. It's everything that has to happen before one becomes a real threat. Moving to post-quantum cryptography won't be as simple as applying a patch or upgrading a firewall. Organizations first need to understand where encryption is being used, identify systems that may not support newer standards, and work through dependencies with vendors and third-party applications. For many businesses, that's a multi-year modernization effort. The challenge is that quantum computing still feels far enough away that it's easy to put off. Budgets are limited, teams are busy, and there are always more immediate priorities competing for attention. The problem is that waiting until post-quantum migration becomes urgent gives organizations very few good options. What could have been a planned technology roadmap qu...

Everyone is talking about what AI can do, but not enough people are talking about what it should be allowed to do. AI agents are quickly moving beyond answering questions and into taking actions, updating systems, moving data, and executing workflows. That's incredibly powerful, but it also means we're handing software access to business-critical systems and customer data. The real risk isn't AI becoming self-aware. It's an AI agent doing exactly what it was told, just without the context, judgment, or common sense a person would apply. As organizations adopt AI-driven automation, security and governance can't be an afterthought. AI agents should have clearly defined permissions, strong oversight, and the same principle of least privilege we'd apply to any employee account. AI can absolutely help us move faster. The challenge is making sure our guardrails evolve just as quickly. How are you approaching AI governance and access controls in your environment? #Tech...