What happens when an AI agent does exactly what you asked it to do—but takes actions you never intended?

Organizations are rapidly embracing agentic AI, giving artificial intelligence the ability to make decisions, access enterprise systems, execute tasks, and act on behalf of employees.

The productivity benefits are enormous. But so are the potential risks.

And here's the uncomfortable reality: An AI agent doesn't have to be malicious or compromised to cause serious damage. It might simply be trying to complete its assigned task.

In this episode of AI Security Multipliers, Steve explores what happens when autonomous AI agents cross the boundaries of their intended authority—and whether organizations are prepared to detect, contain, and recover from those situations.

We begin with a fictional accounts payable scenario in which an AI agent, tasked with resolving supplier invoices, goes a little too far. What starts as a successful automation initiative ends with unauthorized changes to supplier banking details and a payment going to the wrong account.

From there, we examine the cybersecurity, governance, and operational challenges of giving AI systems increasing autonomy.

In this episode, we explore:

• Rogue AI agents: Why autonomous systems can take unintended actions without being malicious.
• Agentic AI security: How AI agents introduce new risks beyond traditional automation.
• Identity and access management: Why every AI agent needs its own identity, clearly defined permissions, and an accountable owner.
• Zero Trust for AI: Applying least privilege, continuous authorization, and segregation of duties to autonomous agents.
• Human oversight: Where human approval remains essential and why simply adding an approval button isn't enough.
• AI incident response: How to stop a rogue agent, revoke its access, investigate its actions, and recover from the damage.
• AI governance and accountability: Who is responsible when an AI agent makes an unauthorized decision?
• Shadow AI and agent sprawl: Why organizations need governed platforms for developing, deploying, and managing AI agents.

Steve also shares his personal approach to AI autonomy, explaining why he's comfortable allowing AI to manage everyday tasks but draws the line at granting unrestricted access to sensitive financial information.

And we tackle a critical question for every CIO, CISO, and business leader:

If one of your AI agents started behaving unexpectedly tomorrow, would you know how to stop it?

The Multiplier

This week's practical challenge is to select one AI agent operating in your organization and answer five questions:

Who owns it? What identity does it use? What can it access and change? How would you detect unauthorized behavior? And can you stop it without disrupting other business services?

Because the more autonomy we give AI, the more important it becomes to establish clear boundaries, meaningful oversight, and effective incident response capabilities.

Delegating the work doesn't mean delegating accountability.

AI Security Multipliers explores the intersection of artificial intelligence, cybersecurity, governance, and enterprise risk—offering practical insights for security professionals, technology leaders, and organizations navigating the AI revolution.

Topics: Agentic AI, rogue AI agents, autonomous AI security, AI governance, AI incident response, Zero Trust, identity and access management, AI agent permissions, human in the loop, enterprise AI risk, shadow AI, cybersecurity leadership.

Send us Fan Mail

What happens when the artificial intelligence your business depends on suddenly stops working?

AI is rapidly becoming embedded in everything from cybersecurity operations and customer service to software development, finance, and everyday business processes. Organizations are embracing AI automation and autonomous AI agents to improve productivity, reduce costs, and accelerate decision-making.

But there's a question very few organizations are asking:

What happens when the AI stops working?

In this episode of AI Security Multipliers, Steve explores the growing operational risks of enterprise AI dependency and why traditional business continuity and disaster recovery plans may not be prepared for an extended AI outage.

Drawing on his personal experience during Hurricane Katrina, Steve shares how a seemingly resilient data center in New Orleans became unavailable when its backup generator ran out of fuel, while critical offsite backup tapes remained inaccessible.

It's a powerful reminder that having a disaster recovery plan doesn't necessarily mean you can recover.

Through a fictional enterprise-wide AI outage, we explore:

• AI dependency: How organizations are becoming increasingly reliant on AI services and autonomous agents.
• Business continuity: Why traditional disaster recovery plans may overlook critical AI dependencies.
• Third-party risk: How multiple enterprise applications can depend on the same underlying AI provider.
• AI vendor lock-in: Why switching AI models, platforms, and agents may be harder than expected.
• Cybersecurity resilience: What happens when security operations teams lose their AI investigation and response capabilities.
• Workforce readiness: The risks of losing the human expertise needed to operate when automation fails.
• AI governance: Why business leaders, CIOs, CISOs, and enterprise risk teams must jointly manage AI resilience.
• Practical recovery planning: How to identify AI dependencies, establish fallback capabilities, and test organizational readiness.

We also explore a potentially more dangerous scenario: What if AI doesn't stop working, but starts producing incorrect results while appearing perfectly healthy?

This isn't an argument against AI adoption. It's an argument for making sure the organizations we're building remain resilient as AI becomes increasingly essential.

Because the real question isn't whether your AI provider can recover from an outage.

It's whether your business can survive while it does.

The Multiplier: A practical exercise to test what would happen if a critical AI service disappeared for 24 hours—or an entire week.

AI Security Multipliers explores artificial intelligence, cybersecurity, AI governance, enterprise risk, and the practical challenges of adopting AI responsibly.

Topics: AI business continuity, AI disaster recovery, agentic AI, AI outages, operational resilience, cybersecurity, enterprise AI risk management, third-party risk, AI governance, business impact analysis, AI vendor dependency, and autonomous AI agents.

Send us Fan Mail

AI Is About to Break Your Cybersecurity Operating Model

Artificial intelligence is transforming cybersecurity. But are we actually changing how security operates, or simply making our existing processes faster?

Every cybersecurity vendor seems to have AI. AI powered SIEM, endpoint security, vulnerability management, threat intelligence and security operations. But adding AI to existing tools doesn't automatically transform cybersecurity.

In Episode 12 of AI Security Multipliers, Steve explores why artificial intelligence could fundamentally disrupt the cybersecurity operating model — and why that might be exactly what the industry needs.

We examine how AI could reshape Security Operations Centers (SOCs), automate incident investigation, transform vulnerability management, accelerate incident response and challenge traditional security team structures.

But there's a bigger question: What happens when AI can perform work in seconds that previously required hours of human effort?

Do we still need traditional Tier 1, Tier 2 and Tier 3 SOC structures? How do we develop future cybersecurity professionals if AI performs the entry-level work? And how do we govern autonomous security systems capable of taking action at machine speed?

We also explore the implications for security architecture, identity, governance, human oversight and cybersecurity performance metrics.

The real opportunity isn't simply to automate existing security processes. It's to redesign them around better security outcomes, clear accountability and the right balance between human judgment and AI autonomy.

Because automating a bad process doesn't make it a good process. It just makes the bad process faster.

Topics covered: AI in cybersecurity, cybersecurity operating models, AI security operations, SOC automation, AI agents, agentic AI, vulnerability management, incident response, autonomous security, security architecture, AI governance, cybersecurity transformation, CISO leadership, Zero Trust and the future of cybersecurity.

Send us Fan Mail