AI-Driven Exposure Management Explained: How to Outpace AI Exploitation in 2026

When AI shrinks the exploit window to hours, exposure management must run at machine speed. See how Bitsight ranks what attackers will hit first in 2026.

AI-driven exposure management uses AI and threat intelligence to continuously discover, prioritize, validate, and remediate exposures at the speed attackers now exploit them. This guide defines the discipline itself, explains how it differs from traditional vulnerability management and Continuous Threat Exposure Management (CTEM), and shows exactly how security teams can operationalize it. You will find a comparison table, a core capabilities breakdown, a five-step operating model mapped to the CTEM lifecycle, real-world examples drawn from CISA's Known Exploited Vulnerabilities (KEV) catalog and Bitsight Threat Intelligence (TI) data, and a detailed look at how Bitsight's platform brings every element together. This page is intended to define the category; for a deeper exploration of what frontier AI means for exposure management programs, and for guidance on how vulnerability management must change in the AI era, see those companion guides.

What is AI-Driven Exposure Management?

AI-driven exposure management is a security discipline that uses artificial intelligence and continuous threat intelligence to discover, contextualize, prioritize, validate, and remediate exposures across an organization's entire attack surface in near real time. Unlike periodic scanning programs, it operates as a continuous loop, ingesting signals from internet-facing assets, underground forums, exploit marketplaces, and dark web channels simultaneously, then using AI models to rank what needs attention first. Bitsight has built its cyber risk intelligence platform around this discipline, combining external asset discovery, Dynamic Vulnerability Exploit (DVE) Intelligence, underground threat intelligence, and third-party risk monitoring into one unified architecture that reflects how exposed an organization actually is, not just what an internal scanner can see.

The term matters because it draws a clear line between programs that catalog vulnerabilities and programs that predict which ones attackers will weaponize. The former counts; the latter decides. AI-driven exposure management is about the latter.

How is it Different from Vulnerability Management and CTEM?

Traditional vulnerability management, Continuous Threat Exposure Management (CTEM), and AI-driven exposure management share some structural DNA but differ substantially in cadence, signal quality, and scope. Understanding those differences helps security leaders assess where their current programs fall short and where investment will close the gap fastest.

Comparison: Traditional VM vs. CTEM vs. AI-Driven Exposure Management

Gartner introduced CTEM in 2022 to formalize continuous, business-aligned exposure reduction as a framework. AI-driven exposure management extends that framework by replacing human-curated prioritization with machine-speed intelligence drawn directly from adversary behavior. Where CTEM defines what to do and in what order, AI-driven exposure management defines how fast and with what signal quality the program must operate given today's threat environment.

Is it the Same as AI Security?

No. AI security refers broadly to securing AI systems, models, pipelines, and data from attack or misuse. AI-driven exposure management is a distinct discipline that uses AI as the analytical engine to manage an organization's broader cyber exposure. The two categories do intersect at one important point: AI infrastructure itself is now a category of external exposure that must be managed. Bitsight Threat Intelligence (TI) observed a 360% increase in exposed AI-related tooling in 2025, with more than one million exposed services identified, led by platforms such as n8n and Open WebUI. Those exposed services represent an attack surface that AI-driven exposure management must cover, even as it also uses AI to process the threat signals that prioritize remediation across all other asset classes. Confusing the two concepts leads organizations to invest in model security while overlooking the far larger problem of unmanaged, internet-facing AI infrastructure.

Why Do AI-Powered Attackers Break Patch-Cycle Programs?

Patch-cycle programs assume the organization has enough time between disclosure and exploitation to assess, test, and deploy a fix. That assumption is no longer reliable. Public proof-of-concept code and AI-assisted exploit adaptation are compressing the disclosure-to-exploitation window significantly, meaning critical vulnerabilities can no longer be patched on a schedule; relevance-based prioritization is now the baseline requirement, according to findings in the Bitsight State of the Underground 2026.

The operational evidence is consistent with that conclusion. Bitsight TI documented a case in From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse where a Russian-speaking actor used a jailbroken AI coding command-line interface (CLI) to complete a full command-and-control (C2) infrastructure migration in approximately six minutes, performing roughly 11% of the work manually. The AI handled the operational complexity. That compression matters for defenders because it means the window between "vulnerability disclosed" and "vulnerability exploited in a campaign" is no longer measured in weeks.

Bitsight TI also observed that Bitsight data recorded nearly 34 million unique vulnerable IPs in 2025, a 45% year-over-year increase, with more than 166,000 vulnerable entities identified, up 27% from the prior year. At that scale, a manual patch-cycle program cannot process the queue fast enough. The math only works when AI models continuously re-score exposure probability and human analysts focus exclusively on the subset that is actively being weaponized.

Three structural dynamics explain why patch-cycle programs fail under these conditions:

1. Exploit acceleration: AI lowers the cost and skill required to adapt a published proof-of-concept into a working exploit. Attackers who once needed specialized knowledge can now iterate faster using AI coding assistants, some of which are accessed through jailbroken guardrails.
2. Scale of underground activity: Cited in Bitsight's AI Abuse report, approximately 5.1 million underground mentions of Gemini, 1.4 million of ChatGPT, 697,000 of Grok, and 656,000 of Claude in 2025. This volume of AI tool discussion in adversarial communities reflects how deeply AI is embedded in the attacker workflow, not as an occasional tool but as a routine capability.
3. Agentic attack execution: The JADEPUFFER operation documented by Bitsight TI represents an illustrative escalation. An agentic ransomware operation entered through an internet-facing Langflow instance using CVE-2025-3248 and pivoted to a production database with minimal human involvement on the attacker side. Patch-cycle programs would not have closed that window in time.

How Do Teams Focus on the Vulnerabilities AI Will Weaponize First?

Focusing on the right vulnerabilities requires a shift from severity-based triage to exploit-likelihood-based prioritization. CVSS scores are useful for understanding technical severity, but they do not capture whether a given vulnerability is being actively discussed in underground markets, whether proof-of-concept code is circulating, or whether ransomware groups have added it to their targeting list. Those signals are what determine whether a vulnerability will be weaponized first.

Core Capabilities Required

• Continuous external discovery including shadow IT and AI assets: The attack surface now includes public LLM (large language model) endpoints, Model Context Protocol (MCP) servers, AI workflow orchestrators, and other components that appear outside of traditional asset inventories. Bitsight TI's watch list for exposed AI infrastructure includes Langflow, Flowise, Open WebUI, Ollama, Gradio, ComfyUI, Jupyter, and MLflow. Each represents a category of internet-facing tooling that creates inherited exposure when unmanaged.

• Exploit-likelihood prioritization from underground chatter: Static CVSS scores do not reflect adversary intent. Prioritization that draws on underground forum activity, dark web marketplace listings, exploit kit mentions, and ransomware group targeting produces a fundamentally different rank order than severity-only approaches. Bitsight's DVE (Dynamic Vulnerability Exploitability) Intelligence scores each CVE against a 90-day exploitation forecast derived from exactly these sources.

• Third-party exposure coverage: A CVE in a key supplier, managed service provider, or software platform can create inherited exposure that extends well beyond an organization's own perimeter. Third-party risk management must extend the same continuous discovery and prioritization logic across vendor and supply chain ecosystems to avoid blind spots at the organizational boundary.

• Remediation validation: Confirming that a patch was applied is not the same as confirming that an exposure is closed. Validation must verify that the remediation actually reduced attack surface, particularly for complex environments where misconfiguration can reopen exposures that were technically patched.

• Board-level metrics: Executives and boards require quantified, benchmarkable evidence of resilience, not vulnerability counts. AI-driven exposure management must translate technical findings into business-risk language, showing trend lines, peer benchmarks, and risk reduction over time in terms that inform investment decisions.

The Five-Step AI-Driven Exposure Management Operating Model

This operating model maps directly to the five CTEM stages established by Gartner: scoping, discovery, prioritization, validation, and mobilization. Each stage is augmented in the AI-driven version with machine-speed intelligence that CTEM's original formulation did not anticipate.

Step 1: Scoping

Define the exposure boundary in business terms, not just network terms. This means identifying critical business processes, the assets and third parties that support them, and the AI infrastructure components (such as public LLM endpoints and MCP servers) that are now part of the operational footprint. Scoping poorly is the most common reason exposure programs miss their highest-risk assets. MCP servers in particular can expose files, environment variables, internal APIs, and shell-backed tools to AI agents; CVE-2025-68143 in mcp-server-git is an example of how a misconfigured AI integration creates a critical exposure that traditional scope definitions would miss entirely.

Step 2: Discovery

Run continuous external discovery across all asset classes within scope, including subsidiaries, cloud services, SaaS applications, shadow IT, and AI infrastructure. Discovery must be agent-free and externally oriented, reflecting what an attacker can see, not just what internal scanners report. Bitsight's internet-scale asset mapping ingests millions of externally observable signals to produce a continuously updated asset inventory.

Step 3: Prioritization

Apply AI-driven exploit likelihood scoring to rank discovered exposures. This step replaces CVSS-only triage with a score that incorporates underground forum chatter, active exploitation data, ransomware targeting signals, and proof-of-concept availability. The goal is a short, defensible list of exposures the team must address this sprint, rather than a queue of thousands with no clear rank order. All 10 of the highest-impact KEVs from 2025 carried Bitsight DVE and CVSS scores above 9, with telecommunications and IT/services sectors most affected and the United States the most exposed country, according to the Bitsight State of the Underground 2026.

Step 4: Validation

Confirm that flagged exposures are actually reachable and exploitable in the organization's specific configuration before committing remediation resources. Validation prevents both over-investment in theoretical risks and under-investment in exposures that appear low-severity but are accessible from the internet.

Step 5: Mobilization

Translate validated findings into ticketed remediation actions with clear ownership and timelines. Mobilization is where exposure management programs most often stall. AI-driven programs accelerate this step by auto-attributing CVEs to affected assets, pre-mapping findings to existing ticketing and workflow systems, and providing remediation guidance in operational terms rather than raw vulnerability data.

Real-World Examples: KEV Data and Bitsight Research

The following examples illustrate how the gap between disclosure and exploitation plays out in practice and what AI-driven exposure management must do to close it.

The CISA Known Exploited Vulnerabilities catalog grew by 245 entries in 2025, reaching 1,484 confirmed exploited vulnerabilities, a 20% year-over-year expansion. CISA's catalog is curated evidence of exploitability, meaning every entry represents a vulnerability that attackers are actively using. Managing exposure against this catalog requires knowing which entries affect your specific assets and which of those are being targeted in campaigns relevant to your sector.

Bitsight TI analysis of the 10 highest-impact 2025 KEVs found that all carried both Bitsight DVE and CVSS scores above 9, with telecommunications and IT/services organizations most heavily affected. The United States was the most exposed country across these entries. This finding confirms that DVE scoring aligns with real-world exploitation at the most critical end of the severity spectrum.

The JADEPUFFER operation documented in the Bitsight AI report From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse shows the speed at which an exposed AI tool becomes an entry point. An agentic ransomware operation entered an environment through an internet-facing Langflow instance via CVE-2025-3248 and subsequently pivoted to a production database. Langflow is on Bitsight TI's exposed AI infrastructure watch list precisely because its internet-facing deployments represent a high-probability initial access vector.

On the credential side, cited in Bitsight's AI Abuse report, 2.8 billion unique compromised credential sets in 2025, with credential volumes growing at a 92% compound annual rate from 2022 to 2025. Compromised credentials feed into exposure: they enable attackers to authenticate rather than exploit, bypassing vulnerability prioritization programs entirely. Identity intelligence and compromised credential monitoring are therefore required components of a complete exposure management program, not optional supplements.

What to Look for in an AI-Driven Exposure Management Platform

Selecting the right platform requires evaluating against a consistent set of capability criteria. Tools from vendors such as Tenable, Qualys, Rapid7, Wiz, and Censys each address portions of the exposure management problem. The question for security leaders is not whether a tool covers one stage but whether it unifies the full operating model and brings exploit-likelihood intelligence, external discovery, and third-party coverage into a single, continuously updated picture.

Must-Have Features for AI-Driven Exposure Management

• Exploit-likelihood scoring with a defined time horizon: A score that identifies "high risk" without specifying the exploitation probability over a defined period is not operationally useful. Look for predictive scoring that provides a concrete time horizon, such as a 90-day exploitation forecast, so teams can align remediation to sprint and patch cycles.
• Underground and dark web intelligence integration: Prioritization signals must originate from where attackers actually operate, not just from public disclosure databases. Platforms that aggregate only National Vulnerability Database (NVD) data will always lag behind programs that monitor underground forums and exploit marketplaces continuously.
• External, agent-free asset discovery: Internal scanner dependency creates gaps wherever agents are not deployed. Effective platforms discover assets as an attacker would, from the outside in, including cloud services, shadow IT, subsidiaries, and AI infrastructure that has not been formally inventoried.
• Third-party and supply chain coverage: An exposure management program that covers only the first-party perimeter misses a growing proportion of enterprise risk. The platform must extend continuous monitoring and exploit-likelihood scoring to vendors, MSPs, and software supply chain components.
• Remediation validation: Discovery and prioritization without confirmation that remediation actually reduced exposure is an incomplete loop. Validation capabilities confirm that the attack surface was genuinely reduced, not just that a ticket was closed.
• Board-ready metrics and benchmarking: Security programs must communicate risk in business terms. Platforms that produce only technical output, such as CVE lists or scan reports, require manual translation before they can inform executive or board decisions. Native quantification and peer benchmarking capabilities close that gap.

How Enterprise Security Teams Apply AI-Driven Exposure Management

Security and risk leaders at large enterprises apply AI-driven exposure management across several distinct use cases. Each one maps to a specific product capability and produces a measurable outcome:

• Attack surface discovery for AI infrastructure: Teams responsible for securing AI-native environments use continuous external discovery to identify which AI tools in their environment are internet-facing, whether those tools carry known CVEs, and whether those CVEs appear on active exploitation lists. This is the use case that most directly addresses the 360% growth in exposed AI-related tooling, cited in Bitsight's AI Abuse report, in 2025.
• CVE prioritization for lean security teams: Organizations with limited remediation capacity use DVE Intelligence to reduce the queue of active CVEs to the small subset with a high 90-day exploitation likelihood. Instead of triaging thousands of AI-surfaced vulnerabilities manually, security teams use DVE to rank them by predicted exploitation likelihood and focus remediation on the small subset likely to be weaponized, according to Bitsight's DVE documentation.
• Third-party exposure monitoring: Supply chain risk teams extend the same exposure discovery and scoring logic to vendors and MSPs, surfacing inherited CVE exposure before it becomes a breach vector. The Bitsight State of the Underground 2026 documents how a CVE in a key supplier or software platform can create inherited exposure outside an organization's own perimeter, an exact use case for supply chain exposure management.
• Credential exposure correlation: Security operations center (SOC) teams correlate compromised credential data with asset inventories to identify accounts that could enable attackers to bypass vulnerability-based controls. With 2.8 billion unique compromised credentials observed in 2025, credential exposure monitoring is not separable from vulnerability exposure management.
• Board and executive reporting: CISOs use benchmarked security ratings and risk quantification outputs to translate technical exposure data into board-ready narratives. Security ratings provide an objective, externally observable measure of exposure that boards can benchmark against industry peers without requiring deep technical context.
• Integrated third-party risk workflows: Risk teams managing vendor portfolios use integrated third-party risk management software to ensure that the same exposure intelligence informing first-party remediation decisions also drives vendor risk assessments, onboarding reviews, and contractual security requirements.

Bitsight differentiates from single-capability tools in this space by delivering all of the above use cases from a single platform with a shared data layer. The DVE score that prioritizes first-party CVEs also scores third-party vendor vulnerabilities. The underground threat intelligence that feeds exploit-likelihood scoring also surfaces compromised credentials and ransomware targeting signals. That architectural coherence eliminates the data inconsistencies that arise when teams stitch together point solutions from multiple vendors.

Best Practices and Expert Tips for AI-Driven Exposure Management

Bitsight's experience working with enterprise security programs surfaces several practices that consistently differentiate high-performing teams from those still struggling with prioritization overload.

1. Reframe the prioritization question: Stop asking "which vulnerabilities are critical?" and start asking "which vulnerabilities will attackers use in the next 90 days?" The first question produces a long list sorted by CVSS. The second produces a short list sorted by exploit likelihood, which is the list remediation teams can actually act on within a sprint cycle.
2. Scope AI infrastructure explicitly: Add AI tools to asset inventories as a named category. Bitsight TI's watch list, including Langflow, Flowise, Open WebUI, Ollama, Gradio, ComfyUI, Jupyter, and MLflow, provides a starting checklist for discovery. Any of these tools exposed to the internet without authentication controls represents a high-probability initial access vector.
3. Treat third-party CVEs as first-party risk: A supplier's unpatched vulnerability is your exposure. Apply the same exploit-likelihood scoring to MSPs, software vendors, and critical suppliers that you apply to internal assets. The guide to TPRM provides a framework for structuring that extension.
4. Monitor underground sources continuously: Prioritization signals from dark web forums and exploit marketplaces lead public disclosure by hours or days in some cases. A program that relies only on NVD publication dates will consistently be late to the most time-critical remediation decisions.
5. Align remediation velocity to threat actor speed: The case documented by Bitsight TI in which a jailbroken AI coding CLI completed a C2 migration in approximately six minutes establishes a benchmark for attacker operational speed. Remediation programs measured in weeks are structurally misaligned with that reality. Continuous monitoring combined with pre-built remediation workflows compresses the mean time to remediation to a timeframe that is operationally meaningful.
6. Use board metrics to sustain program investment: Security programs that cannot demonstrate risk reduction in quantified, benchmarkable terms struggle to maintain budget during economic compression cycles. Exposure management programs that produce board-ready metrics survive those cycles. Security posture management capabilities that combine exposure data with control governance and benchmarking provide the output format that sustains executive confidence.
7. Validate, do not just close tickets: Confirmed remediation is the only closure that reduces exposure. Build validation steps into every remediation workflow and track confirmed exposure reduction as the primary program metric, not ticket closure rate.

Advantages and Benefits of AI-Driven Exposure Management

AI-driven exposure management delivers measurable operational advantages over both traditional VM programs and CTEM programs that have not incorporated machine-speed intelligence.

• Shorter mean time to prioritize: By replacing manual CVE triage with AI-scored exploit-likelihood rankings, teams move from days of analyst effort per vulnerability batch to near-real-time rank ordering. This compression is most significant for organizations managing large CVE queues across complex environments.
• Higher remediation precision: When prioritization is informed by actual underground activity rather than theoretical severity, remediation resources concentrate on exposures that are being actively weaponized. This reduces wasted effort on vulnerabilities that carry high CVSS scores but low real-world exploitation probability.
• Extended coverage without proportional headcount: Continuous external discovery and AI-driven scoring extend effective coverage to cloud assets, subsidiaries, third parties, and AI infrastructure without requiring proportional increases in analyst headcount. The platform scales the intelligence layer; analysts scale the decision layer.
• Reduced inherited third-party risk: Programs that extend exposure discovery and exploit-likelihood scoring to vendors and suppliers surface third-party CVE exposure before it creates a breach pathway, closing the blind spot that patch-cycle-only programs leave at the organizational boundary.
• Board-level defensibility: Quantified, benchmarkable exposure metrics give CISOs and risk officers the evidence base to justify security investment in business terms, demonstrating risk reduction over time rather than relying on vulnerability counts that boards cannot contextualize.
• Continuous posture improvement: Because the operating model is a loop rather than a project, exposure management programs built on AI-driven intelligence improve continuously. Each remediation cycle feeds back into the discovery and prioritization stages, producing a progressively more accurate and current picture of organizational exposure.

How Bitsight Operationalizes AI-Driven Exposure Management

Bitsight's platform is architecturally designed for the AI-driven exposure management operating model described in this guide. Five product capabilities map directly to the five operating model steps.

• Attack Surface Intelligence (ASI): Bitsight's external attack surface management capability extends EASM by adding real-time threat intelligence from the deep, dark, and clear web, so security teams know not just what is exposed but which exposures are actively being targeted. ASI provides the continuous, agent-free external discovery that scoping and discovery stages require, covering domains, IPs, cloud services, shadow IT, and AI infrastructure components.
• DVE Intelligence: Bitsight's Dynamic Vulnerability Exploit (DVE) Intelligence delivers AI-cultivated CVE risk assessments within hours of publication, predicting exploitation likelihood up to 90 days in advance. DVE draws on underground forum chatter, active exploitation data, and ransomware targeting signals to produce a score that complements CVSS with real-world adversary behavior. DVE Intelligence also performs automated CPE (Common Platform Enumeration) to CVE matching, addressing data deficiencies in the NVD's CPE dictionary to improve attribution accuracy.
• Cyber Threat Intelligence (CTI): Bitsight's cyber threat intelligence platform collects 7 million intelligence items daily from more than 1,000 underground forums and marketplaces, processing them through AI enrichment in under one minute. CTI provides the underground signal layer that transforms DVE scoring from a theoretical model into one grounded in observed adversary behavior. Bitsight TRACE, the in-house research team that produced both the Bitsight State of the Underground 2026 and From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse, combines AI-driven analysis with human analyst expertise to ensure that automated scoring is anchored in contextual understanding of actual threat actor operations.
• Security Posture Management (SPM): Bitsight's security posture management platform combines threat intelligence, business context, control governance, and peer benchmarking to produce a board-ready view of organizational resilience. SPM bridges the gap between operational exposure data and executive communication, ensuring that the risk reduction produced by exposure management programs is visible and credible at the governance level.
• Third-Party Risk Management (TPRM): Bitsight's TPRM platform extends the same continuous monitoring and exploit-likelihood intelligence across vendor and supply chain ecosystems. This integration ensures that the exposure management program's scope does not end at the organizational boundary, covering the inherited CVE risk that originates in suppliers, MSPs, and software platforms. The best TPRM platforms guide provides additional context on what to evaluate in this category.

Taken together, these five capabilities form a coherent architecture that covers every stage of the AI-driven exposure management operating model without requiring integration of multiple point solutions. Bitsight has been named a Leader in The Forrester Wave for Cybersecurity Risk Ratings Platforms and a Visionary in the 2026 Gartner Magic Quadrant for Cyber Threat Intelligence Technologies, reflecting the breadth and depth of the platform across the exposure management lifecycle.

Read the Bitsight State of the Underground 2026 to understand the full scope of the threat landscape driving the need for AI-driven exposure management.

Key Takeaways and How to Get Started

AI-driven exposure management is not a product category; it is a discipline that defines how security programs must operate when attackers use AI to compress the time between vulnerability disclosure and active exploitation. The key takeaways from this guide are as follows.

First, patch-cycle programs are structurally misaligned with the current threat environment. Bitsight TI data from 2025 documents nearly 34 million unique vulnerable IPs, a 45% year-over-year increase, across more than 166,000 vulnerable entities. Manual triage cannot process that queue at the speed exploitation now occurs.

Second, AI infrastructure is now an attack surface category in its own right. The 360% increase in exposed AI-related tooling in 2025 means that any organization running internet-facing AI tools without continuous monitoring has an unscoped exposure.

Third, exploit-likelihood scoring from underground sources is the minimum viable prioritization standard for 2026. Static CVSS triage produces lists that do not reflect adversary intent. Programs that have not integrated underground signal into their prioritization logic are operating with degraded intelligence.

Fourth, third-party exposure is first-party risk. Inherited CVE exposure from suppliers and MSPs creates breach pathways that vulnerability management programs scoped only to internal assets cannot detect.

Fifth, board-level metrics are a program requirement, not a reporting nicety. Security programs that cannot quantify risk reduction in executive terms will struggle to sustain investment through business cycles.

Bitsight has built its platform to deliver all five of these capabilities from a unified architecture. To see how it applies to your organization's specific exposure profile, request a Bitsight demo and assess your current program against the demands of the 2026 threat environment. To understand the full scope of AI abuse in the underground threat landscape, read the Bitsight AI report From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse.

FAQs About AI-Driven Exposure Management

What is AI-driven exposure management?

AI-driven exposure management is a security discipline that uses artificial intelligence and continuous threat intelligence to discover, prioritize, validate, and remediate exposures at the speed attackers now exploit them. It extends the CTEM framework introduced by Gartner in 2022 by replacing periodic, manual prioritization with machine-speed scoring derived from underground adversary behavior. Bitsight operationalizes this discipline through Attack Surface Intelligence, DVE Intelligence, cyber threat intelligence, Security Posture Management, and Third-Party Risk Management in a unified platform.

How is AI-driven exposure management different from traditional vulnerability management?

Traditional vulnerability management relies on periodic scans, CVSS severity scoring, and manual triage. AI-driven exposure management operates continuously, uses exploit-likelihood scoring derived from underground forums and active exploitation data, and covers external assets, shadow IT, AI infrastructure, and third parties that internal scanners cannot reach. The operational result is a materially shorter list of exposures requiring urgent remediation and significantly less time spent on vulnerabilities that carry theoretical severity but low real-world exploitation probability. Bitsight's DVE score is specifically designed to produce that shorter, higher-confidence list.

Why do AI-powered attackers break patch-cycle programs?

Patch-cycle programs depend on having days or weeks between vulnerability disclosure and active exploitation. Bitsight TI documented that public proof-of-concept availability and AI-assisted exploit adaptation are compressing that window dramatically. A jailbroken AI coding CLI enabled a threat actor to complete a full C2 infrastructure migration in approximately six minutes with roughly 11% of the work done manually. When attackers operate at that speed, a remediation program measured in patch cycles will consistently be late to the exposures that matter most.

What does it mean that MCP servers are an exposure management concern?

Model Context Protocol (MCP) servers connect AI agents to files, environment variables, internal APIs, repositories, and shell-backed tools. When an MCP server is misconfigured or carries a known vulnerability, it can expose internal systems to AI agents operating on behalf of an attacker. Bitsight TI identified CVE-2025-68143 in mcp-server-git as an example of this risk class. AI-driven exposure management programs must scope MCP servers and other AI integration components as a distinct asset category requiring continuous external discovery and CVE attribution.

How does Bitsight help security teams focus on the vulnerabilities AI will weaponize first?

Bitsight's DVE (Dynamic Vulnerability Exploitability) Intelligence scores each CVE against a 90-day exploitation forecast derived from underground forum chatter, dark web marketplace listings, active exploitation telemetry, and ransomware targeting signals. This score complements static CVSS ratings with real-world adversary behavior, producing a rank-ordered prioritization list that reflects exploitation probability rather than theoretical severity. Security leads can use DVE output to concentrate remediation on the small subset of vulnerabilities most likely to be weaponized, reducing alert fatigue and improving remediation precision without requiring additional analyst headcount.

What exposed AI infrastructure should security teams monitor?

Bitsight TI has identified a watch list of AI infrastructure components that create high-probability exposure when internet-facing and unmanaged: Langflow, Flowise, Open WebUI, Ollama, Gradio, ComfyUI, Jupyter, and MLflow. Each represents a category of AI tooling with known CVE risk and active exploitation history. The JADEPUFFER operation, which entered through an internet-facing Langflow instance via CVE-2025-3248 and pivoted to a production database, illustrates the concrete risk these components carry when they fall outside the discovery scope of an exposure management program.

What role does third-party risk play in AI-driven exposure management?

Third-party risk is integral to AI-driven exposure management because a CVE in a supplier, MSP, or software platform creates inherited exposure outside an organization's own perimeter. Bitsight TI data from the Bitsight State of the Underground 2026 confirms this pattern, documenting how supplier-side vulnerabilities translate into first-party breach risk. Bitsight's TPRM platform extends the same continuous monitoring and exploit-likelihood intelligence across vendor ecosystems, ensuring that the exposure management program's scope does not create blind spots at the organizational boundary. View Bitsight pricing to understand how TPRM integrates with the broader platform.

Last updated: September 29, 2026. Statistics and examples drawn from Bitsight TRACE | State of the Underground 2026 and From Jailbreaks to Agentic Attacks: The Evolution of AI Abuse (August 2026), covering 2025 calendar year threat data observed by Bitsight Threat Intelligence (TI).