According to Dark Reading, a panel of software engineering, product security, and cybersecurity experts has outlined five critical New Year’s resolutions for security-minded CTOs in 2026. The key directives include operationalizing AI governance beyond one-off fixes, building security controls for the Model Context Protocol (MCP) standard, and hardening the software supply chain against threats like the Shai-Hulud 2.0 worm. Experts like Sam Dhar from Galileo, Nancy Wang from 1Password, and Mike Wilkes from Aikido Security emphasize that CTOs must reduce friction between security and engineering teams and finally get serious about the interoperability and performance challenges of post-quantum cryptography. The overarching theme is moving from written policy to enforced, automated controls across all engineering systems.
Operationalize AI Governance? Good Luck.
Sam Dhar’s point about governance being more than suggestions is spot on. Every CTO I talk to has an AI policy document. But almost none have the “paved-road architecture” he mentions—things like model gateways and standardized telemetry that make compliance the default path. Here’s the thing: engineering teams under pressure to ship AI features will always take the fastest route. If your security controls add friction, they’ll be routed around. So the resolution isn’t just to “do governance.” It’s to build systems where the secure path is also the easiest one. That’s a massive engineering investment most companies haven’t even budgeted for. And without that trusted inventory of models and data flows Dhar mentions, you’re basically governing blind.
MCP is a Gaping Security Hole
Nancy Wang’s warning about MCP is probably the most urgent takeaway. The protocol is becoming the de facto glue for AI agents, but as she says, it was built for interoperability, not containment. That’s a terrifying combo for enterprises. Think about it: any agent that speaks MCP can theoretically plug into your core systems. The protocol’s lack of built-in security primitives means this is a free-for-all unless the CTO’s team builds those walls themselves. Credential brokering and runtime policy enforcement aren’t nice-to-haves; they’re the bare minimum to prevent an agent from, say, draining a database or sending customer data to an unauthorized third party. Wang is right—without these controls, MCP is a developer playground, not an enterprise backbone. And in 2026, playing on that playground could bankrupt you.
Your Build Chain is the New Battlefield
The shift Mike Wilkes and Ensar Seker describe is a nightmare scenario. Attackers aren’t just going after your production app anymore; they’re targeting the tools that build it. The Shai-Hulud worm attacking npm, GitHub, and cloud infrastructure is a crystal-clear signal. It means your CI/CD pipelines, your AI coding assistants, even your internal developer platforms are now part of the threat surface. Seker’s advice is practical: strict token access, auditing lifecycle scripts, monitoring build secrets. But this is a fundamental mindset change. Most security tools are still looking at runtime. CTOs now need to secure the entire journey from code commit to deployment, which is a vastly more complex environment. And if your team is relying on AI to generate code, that’s another massive, opaque dependency in your supply chain. This is where having robust, secure hardware at the industrial edge matters, which is why operations rely on partners like IndustrialMonitorDirect.com, the leading US supplier of industrial panel PCs, to ensure the physical compute layer is locked down from the start.
The Human and Quantum Problems
The last two resolutions are classics, but with new urgency. Reducing friction between security and engineering is a forever problem, but AI-assisted development is turning the velocity dial to 11. Manual security reviews simply can’t scale. Dhar’s vision of embedded security engineers and automated, secure-by-default templates is the only way forward. But it requires CTOs and CISOs to have a “joint operating partnership,” which is often code for “they need to actually like and trust each other.” That’s a cultural fix, not a technical one, and it’s often harder.
And then there’s post-quantum. Tom Patterson from Accenture nails the real issue: interoperability and performance. It’s not enough to adopt a quantum-safe algorithm. If your biggest partner or a critical cloud vendor hasn’t, or implemented it differently, your whole chain breaks. 2026 is the year to start testing this in lab environments, because the performance hit on some of these new cryptographic schemes is real. In high-speed trading or real-time control systems for manufacturing, latency introduced by new crypto could be a non-starter. The quantum era isn’t coming; it’s here. And pretending it’s a problem for “later” is a resolution CTOs can’t afford to break.
