What we're learning from financial services customers about the defensive context AI needs to make better security decisions.

We've been spending time with customers across financial services, including large banking institutions, discussing how they're thinking about AI, where they see opportunities to improve security operations, and what needs to change before AI agents can take on more meaningful work.

These organizations operate some of the most complex security environments in the world, with significant investments in technology, sophisticated defensive capabilities, and experienced security teams. Yet a common challenge continues to surface: understanding how all those technologies work together to defend against the adversaries that matter to the business.

One reason is organizational silos. In large, sophisticated security programs, individual teams have deep expertise in the technologies and domains they manage, but often lack visibility into how those capabilities work together. The challenge isn't a lack of expertise. It's connecting that expertise across teams to build a shared understanding of the organization's overall defensive posture.

It's a difficult question even without AI. Individual technologies provide visibility into what they detect, prevent, or monitor, but bringing that information together to understand the effectiveness of the defense as a whole takes considerable effort.

Now, as organizations explore how AI agents can analyze security information, automate operational workflows, and accelerate decisions, that challenge takes on new significance.

What does an AI agent need to understand about an organization's defenses before it can recommend what to do next?

AI Can Analyze Security Data. Can It Understand the Defense?

Consider a straightforward request to an AI agent: Which defensive gap should we address first?

An agent can analyze vulnerability reports, compare severity scores, summarize threat intelligence, and organize findings quickly. Those capabilities can save security teams considerable time, but they don't necessarily produce the right recommendation.

To answer that question meaningfully, an agent needs to understand which adversaries are relevant to the organization and how they operate. It also needs to know which defensive technologies are deployed, what behaviors those technologies can address, where coverage may be insufficient, and what testing has confirmed about those weaknesses and the defenses intended to address them.

This is especially important in complex environments where multiple technologies may address the same adversary technique without necessarily covering the specific procedure an attacker uses.

An apparent gap might already be addressed by an existing technology that needs better configuration. A vulnerability with a high severity score may represent less immediate risk than a defensive weakness affecting a critical business system. And a recommendation to purchase another product might overlook capabilities the organization already owns.

Without a connected understanding of the environment, AI can prioritize the most visible finding rather than the most consequential one. Because AI can generate convincing explanations from incomplete information, those recommendations may sound more authoritative than the evidence supports.

The opportunity isn't simply to help AI process more security data. It's to give AI the defensive context it needs to make that data useful.

What We're Learning From Financial Services Customers

One of the more interesting themes emerging from our customer conversations is how organizations want to make defensive information available beyond traditional security dashboards and reporting.

Today, understanding defensive coverage often requires bringing together information from multiple technologies, interpreting how those capabilities relate to adversary behavior, and determining where meaningful gaps may exist. For large financial institutions, where security environments have grown increasingly complex, connecting those pieces can require considerable manual effort.

Customers are looking for ways to make that understanding more accessible and actionable, particularly as they explore how AI could support more of their security operations.

These discussions also suggest possibilities around APIs, Model Context Protocol (MCP), and agent-based workflows that could allow shared defensive context to support other applications and systems.

Imagine an AI agent examining an emerging adversary procedure and identifying where existing defenses may need attention. Another workflow could evaluate defensive gaps affecting a critical business application, drawing on the same underlying information rather than requiring a new manual assessment.

An agent could also help analyze changes in adversary behavior, connect those changes to existing defensive capabilities, and prepare recommended actions for human review.

These are potential applications, not capabilities customers have necessarily deployed at scale. But they illustrate an important shift in how organizations are thinking about defensive posture information.

Rather than treating defensive coverage as something people periodically assess through a dashboard, organizations are beginning to explore how that context could become part of the systems and workflows where security decisions happen.

Tidal Cyber helps provide that foundation by connecting adversary behaviors to the defensive capabilities organizations already have in place. That threat-led understanding of coverage can inform how people, applications, and AI agents evaluate gaps and determine what needs attention.

Access to more security data alone doesn't ensure better recommendations from an AI agent. Connecting that information to the threats an organization faces and the defenses available to address them can help an agent make better recommendations.

AI May Accelerate Decisions, but the Adversary Isn't Standing Still

There's another reason these conversations are becoming more relevant.

Much of the discussion around AI in cybersecurity focuses on how attackers can automate reconnaissance, experiment with techniques, or scale their operations. While the underlying adversary behaviors aren't necessarily new, AI and automation can help attackers execute familiar procedures faster and adapt more frequently.

For defenders, that places greater pressure on the time between learning something new about an adversary and determining whether existing defenses can address it.

Responding isn't always straightforward. A change in adversary behavior can affect multiple technologies and require significant analysis to understand whether an organization has the right capabilities in place. Even when the intelligence is available, translating it into a defensive decision can take time.

AI has the potential to accelerate that process. To support reliable defensive recommendations, it needs an accurate understanding of the environment.

This is where Threat-Led Defense provides an important foundation. By connecting relevant adversary procedures to existing defensive capabilities, organizations can identify potential gaps, prioritize improvements, and use testing to validate whether changes deliver the intended protection.

AI agents could help accelerate parts of that cycle, analyzing changes in adversary behavior, examining existing coverage, and preparing recommendations that security teams can evaluate and act on.

The goal isn't to remove human judgment from security decisions. It's to reduce the manual effort required to understand what changed, what it means for the defense, and what needs attention.

As adversaries move faster, that ability becomes increasingly valuable.

Building the Foundation for AI-Enabled Security Operations

One of our biggest takeaways from these conversations is that successful AI adoption in security may depend as much on the information organizations make available to agents as on the capabilities of the agents themselves.

Large financial institutions already generate enormous volumes of security information. They have threat intelligence, vulnerability data, detection results, configuration information, and telemetry from numerous technologies. What they don't always have is a consistent way to connect that information to how their defenses work against specific adversary behaviors.

For years, organizations have worked to bring these different views together through assessments, reporting processes, and manual analysis. AI creates an opportunity to change how that work gets done, but it doesn't eliminate the need for a reliable understanding of the defense.

Organizations are beginning to think beyond AI as a tool for summarizing alerts or analyzing findings. They're exploring how AI could help interpret changes in adversary behavior, understand the implications for existing defenses, and support the decisions that follow.

For Tidal, this is a natural extension of Threat-Led Defense. Understanding the adversary, connecting that behavior to defensive capabilities, and identifying where coverage needs improvement provide the context that both people and AI systems can use to make more informed decisions.

The next step is making that understanding available wherever decisions happen, whether that's through an analyst investigating a new threat, an application pulling defensive coverage data through an API, or an AI agent preparing a recommendation for review.

There is still work ahead, and many of these applications remain exploratory. But the direction is becoming clearer. As organizations invest in AI-enabled security operations, the quality of their decisions will depend on more than the sophistication of their models or the volume of data they can process.

Before we ask AI to make more security decisions, we need to make sure it understands the defenses those decisions are meant to improve.

For organizations that have spent years building sophisticated security environments, the opportunity isn't necessarily to add more technology. It's to understand what they already have, how it performs against relevant adversary behaviors, and how to put that knowledge to work.

Frequently Asked Questions

What is defensive context in AI cybersecurity?

Defensive context connects the threats an organization faces with its deployed security capabilities, configurations, critical systems, and evidence from testing. It helps an AI agent assess what a finding means for that organization and recommend where security teams should focus.

What information does an AI agent need to prioritize defensive gaps?

An AI agent needs information about relevant adversaries and their procedures, deployed defenses, configurations, critical business systems, and available validation results. It also needs to distinguish confirmed weaknesses from potential gaps and missing information. Together, these inputs support recommendations that security teams can evaluate.

How does Threat-Led Defense support AI-enabled security operations?

Threat-Led Defense connects relevant adversary behavior to an organization's defensive capabilities. That context can help AI agents assess changes in threats, identify potential coverage gaps, and prepare recommendations for human review. Security teams still need to test whether the recommended changes deliver the intended protection.

Does mapping defensive coverage prove that a security control works?

No. Coverage mapping identifies capabilities that may address an adversary technique or procedure. Demonstrating effectiveness requires evidence from the organization's environment, such as testing, configuration checks, or observed detection and prevention results. AI recommendations should distinguish expected coverage from demonstrated performance.