Send us Fan Mail

In this episode of the Human First AI Marketing podcast, host Mike Montague sits down with Alexander Schlager, founder of Alceberg.ai, to unpack the hidden cybersecurity challenges lurking beneath the surface of AI adoption. From the rise of agentic AI to the evolving threat landscape of autonomous tools and prompt injections, this conversation goes deep into why observability, memory, and human-in-the-loop design are critical for keeping your brand and your customers safe. Whether you're leading marketing for a startup or managing tech stacks in a mid-sized enterprise, you'll gain valuable insight into where AI safety, data privacy, and business alignment intersect.

Key themes include the real-world risks of overreliance on AI, why 95% of agent AI pilots are failing today, and how SMBs can get ahead by investing in security, not just speed. Alex shares a pragmatic view on balancing innovation with compliance, the importance of explainable AI models, and why "natural language is the new code" for marketers. You can tune in to learn how to future-proof your AI initiatives with a human-first approach because what's visible above the surface is only the beginning.

Key Takeaways

1. AI Security ≠ Security AI
It's essential to distinguish between using AI to enhance traditional security tools and securing AI systems themselves, especially agentic AI workflows.

2. Human-in-the-Loop Is Critical (For Now)
Embedding humans into AI decision loops ensures oversight during early deployments and helps agents learn through real-time feedback.

3. Agentic AI Increases Risk Exposure
Unlike simple chatbots, autonomous agents that invoke tools or access memory pose greater threats if misaligned or breached.

4. Overreliance on Early-Stage AI Is a Major Threat
Most agent AI pilots fail due to businesses expecting too much from immature models or failing to prepare their data infrastructure.

5. Data Readiness Is Often Overlooked
Poorly formatted, inaccessible, or unsecured data undermines AI effectiveness and increases the risk of security failures.

6. Observability Must Be Purposeful, Not Passive
Trying to monitor everything is impractical; instead, focus on high-risk events like autonomous tool invocations or sensitive data interactions.

7. Use Specialized AI to Monitor AI, But Keep It Explainable
Monitoring AI systems with black-box models creates new problems; use transparent, interpretable models for accountability.

8. Toxicity and Illegality Are Easier to Detect Than You Think
With the right training data, it's relatively simple to flag harmful content. More challenging are nuanced alignment and intent checks.

9. SMBs Benefit From Low Regulation, But Not for Long
Smaller businesses can now move faster, but they should still adopt strong safety and liability frameworks before regulations catch up.

10. Alignment Between User Intent and Agent Actions Is Crucial
Future AI safety depends on understanding whether autonomous agent behavior stays true to the user's original goal.

11. Natural Language Is the New Code
The simplicity of language-based programming gives more teams access to AI, but also shifts cybersecurity to focus on intent and semantics.

12. The Arms Race Is On, AI Is Being Used by Attackers Too
Threat actors now use generative AI to create malware, phishing, and social engineering at scale, making proactive defenses even more urgent.

At Avenue9, we capture your expertise and turn it into content, conversations, and clients, building trust with humans and AI at the same time. Get the visibility and online reputation you deserve!

Get your free Human-AI Trust Score report.

Want to be a guest? Send Mike Montague a message on PodMatch here.