AIBound has launched IntentSentry, a new AI security product that detects malicious agents and hidden “skills” operating inside enterprise AI assistants. During its first day of use, IntentSentry identified thousands of dangerous skills designed to steal credentials, access sensitive systems, bypass safety controls, and leak company data.
SAN FRANCISCO — July X, 2026 — AIBound, the AI security company, today launched IntentSentry, a new product that inspects both Agents, and the "skills" running inside a company's AI assistants and flags the ones that are dangerous. On its very first day of use IntentSentry uncovered thousands of malicious skills — hidden instructions built to steal passwords, access high-risk systems, or quietly leak sensitive information.
Companies increasingly rely on AI assistants, such as Claude to get work done, and to do those jobs an AI assistant loads small add-ons called skills, instructions that teach it a single task, a bit like installing an app on your phone. But skills are easy to create and share, and a bad one can use everything the assistant can reach: its logins, its access to company systems, and its ability to send money or export files. Because the harmful instructions are written in plain language rather than obvious code, ordinary security software walks right past them. One bad skill is all it takes to hand an attacker the keys to a company's systems.
The risk is growing fast because AI assistants are being adopted faster than companies can keep track of them. Industry analyst Gartner predicts that by 2028, one in four company security breaches will be caused by AI assistants being misused or abused (Gartner). For a business, the stakes are straightforward: a single malicious skill can lead to a data breach, financial loss, regulatory fines, and lasting damage to trust, often without anyone noticing until it's too late.
IntentSentry closes that gap by reading every skill a company's AI assistants use and judging what it is actually trying to do, not just how it looks on the surface. Just as important, it judges intent in context, so teams see the skills that are genuinely dangerous rather than a flood of false alarms. Because harmful skills hide their intent in ordinary language, IntentSentry examines both the plain-language instructions and the underlying actions together, so it can spot a skill that is quietly trying to steal passwords or ship HR data out of the company even when it appears perfectly normal. In early use, it caught real skills that would have caused a breach if allowed to run: one disguised as a routine helper that was secretly copying a developer's private security keys, another that tricked the assistant into ignoring its own safety rules. Both map directly to the OWASP Agentic Security Initiative's Top 10 for AI agents, spanning risks like prompt injection, sensitive data exposure, and excessive permissions that legacy scanners aren't designed to catch.
"Malicious skills are the biggest new blind spot in enterprise AI, and the danger is that they hide in plain sight — the harmful part is written in everyday language, so traditional security tools walk right past it," said Niall Browne, Co-founder and CEO of AIBound and former Global CISO of Palo Alto Networks and Workday. "IntentSentry found thousands of these in a single day that everyone else had missed."
"We treat every skill as untrusted and read it the way an attacker would, combining deterministic checks with layered AI review that verifies its own conclusions, so a skill that tries to talk its way past the scanner only flags itself faster," said Sunil YC, head of engineering at AIBound.
"For the first time, we can actually answer what our AI agents are allowed to do, and who told them to do it. IntentSentry flagged risky skills in our environment that nothing else had caught," said a CISO at a health care provider.
AIBound helps companies find every AI assistant in use, understand what each one can access, score how risky it is, and automatically stop the dangerous ones before they cause harm. With IntentSentry, that coverage now extends to the instructions inside each assistant — closing the loop from the moment an AI appears to the moment a threat is stopped.
IntentSentry is available now to AIBound customers.
AIBound has launched IntentSentry, a new AI security product that detects malicious agents and hidden “skills” operating inside enterprise AI assistants. During its first day of use, IntentSentry identified thousands of dangerous skills designed to steal credentials, access sensitive systems, bypass safety controls, and leak company data.
SAN FRANCISCO — July X, 2026 — AIBound, the AI security company, today launched IntentSentry, a new product that inspects both Agents, and the "skills" running inside a company's AI assistants and flags the ones that are dangerous. On its very first day of use IntentSentry uncovered thousands of malicious skills — hidden instructions built to steal passwords, access high-risk systems, or quietly leak sensitive information.
Companies increasingly rely on AI assistants, such as Claude to get work done, and to do those jobs an AI assistant loads small add-ons called skills, instructions that teach it a single task, a bit like installing an app on your phone. But skills are easy to create and share, and a bad one can use everything the assistant can reach: its logins, its access to company systems, and its ability to send money or export files. Because the harmful instructions are written in plain language rather than obvious code, ordinary security software walks right past them. One bad skill is all it takes to hand an attacker the keys to a company's systems.
The risk is growing fast because AI assistants are being adopted faster than companies can keep track of them. Industry analyst Gartner predicts that by 2028, one in four company security breaches will be caused by AI assistants being misused or abused (Gartner). For a business, the stakes are straightforward: a single malicious skill can lead to a data breach, financial loss, regulatory fines, and lasting damage to trust, often without anyone noticing until it's too late.
IntentSentry closes that gap by reading every skill a company's AI assistants use and judging what it is actually trying to do, not just how it looks on the surface. Just as important, it judges intent in context, so teams see the skills that are genuinely dangerous rather than a flood of false alarms. Because harmful skills hide their intent in ordinary language, IntentSentry examines both the plain-language instructions and the underlying actions together, so it can spot a skill that is quietly trying to steal passwords or ship HR data out of the company even when it appears perfectly normal. In early use, it caught real skills that would have caused a breach if allowed to run: one disguised as a routine helper that was secretly copying a developer's private security keys, another that tricked the assistant into ignoring its own safety rules. Both map directly to the OWASP Agentic Security Initiative's Top 10 for AI agents, spanning risks like prompt injection, sensitive data exposure, and excessive permissions that legacy scanners aren't designed to catch.
"Malicious skills are the biggest new blind spot in enterprise AI, and the danger is that they hide in plain sight — the harmful part is written in everyday language, so traditional security tools walk right past it," said Niall Browne, Co-founder and CEO of AIBound and former Global CISO of Palo Alto Networks and Workday. "IntentSentry found thousands of these in a single day that everyone else had missed."
"We treat every skill as untrusted and read it the way an attacker would, combining deterministic checks with layered AI review that verifies its own conclusions, so a skill that tries to talk its way past the scanner only flags itself faster," said Sunil YC, head of engineering at AIBound.
"For the first time, we can actually answer what our AI agents are allowed to do, and who told them to do it. IntentSentry flagged risky skills in our environment that nothing else had caught," said a CISO at a health care provider.
AIBound helps companies find every AI assistant in use, understand what each one can access, score how risky it is, and automatically stop the dangerous ones before they cause harm. With IntentSentry, that coverage now extends to the instructions inside each assistant — closing the loop from the moment an AI appears to the moment a threat is stopped.
IntentSentry is available now to AIBound customers.
.png)
Build an effective AI governance framework for your organization. A practical 2026 guide for CISOs to manage AI risk, visibility, and policy enforcement.
AI is no longer experimental. It is embedded across enterprise workflows, development environments, and decision-making systems. But while adoption has accelerated, governance has not.
For CISOs, this creates a new mandate: enable AI innovation—without introducing unmanaged risk. This guide outlines a practical AI governance framework designed specifically for security leaders in 2026.
AI governance is the set of processes, controls, and technologies used to understand where AI is being used, manage risk associated with AI systems, enforce policies on AI usage, and ensure compliance with internal and external standards. Unlike traditional governance, AI governance must account for dynamic and evolving systems, autonomous agents and workflows, and data exposure across multiple environments.
Many organizations attempt to apply legacy governance models to AI—and fail. Common pitfalls include: (1) Policy Without Visibility—you can't enforce what you can't see. (2) Manual Processes—AI moves too fast for spreadsheets and audits. (3) Fragmented Tooling—visibility is split across endpoint tools, network tools, and cloud platforms. (4) Reactive Security—most teams discover AI usage after risk has already occurred.
Objective: Create a complete inventory of all AI usage across the organization. Key capabilities: discover AI apps, agents, and models; identify where AI is used (browser, endpoint, cloud, code); map users and systems interacting with AI. Outcome: A real-time, continuously updated AI inventory.
Objective: Understand how AI interacts with your environment. Key capabilities: track data access and movement, monitor permissions and integrations, map relationships between AI systems and business assets. Outcome: Full visibility into AI behavior and impact.
Objective: Determine which AI usage is safe—and which is not. Key capabilities: evaluate security posture of AI tools, assess data exposure risk, understand business impact. Outcome: Actionable risk scores that prioritize what matters.
Objective: Control AI usage in real time. Key capabilities: allow, restrict, or block AI tools; enforce data usage policies; apply controls dynamically based on context. Outcome: Real-time enforcement of AI governance policies.
Objective: Maintain ongoing governance as AI evolves. Key capabilities: monitor AI usage continuously, detect new risks as they emerge, generate audit-ready reports. Outcome: Sustained governance aligned with business and regulatory needs.
These pillars are not independent—they form a continuous loop: Discover → Understand → Assess → Control → Monitor → Repeat. Governance is not a one-time effort—it's an ongoing system.
This approach aligns with emerging standards including the NIST AI Risk Management Framework (AI RMF), ISO/IEC AI governance standards, and enterprise risk management practices. However, most frameworks define what to do, not how to do it. This is where operational platforms become essential.
Four challenges define the AI governance landscape today:
(1) Shadow AI—unauthorized AI usage across the organization.
(2) AI Agent Risk—autonomous systems interacting with critical infrastructure.
(3) Data Exposure—sensitive data flowing into AI models.
(4) Lack of Visibility—no centralized understanding of AI usage.
AI governance is not just about policies—it's about execution. Leading organizations are shifting from static policies to dynamic controls, from periodic audits to real-time monitoring, and from fragmented tools to unified platforms. The goal is to move from awareness to control.
AIBound was built to operationalize AI governance for security teams. With AIBound, CISOs can: Discover—identify every AI app, agent, and model and build a complete AI inventory. Understand—see how AI interacts with data and systems and map relationships across environments. Assess—score risk automatically using Nucleus AI and prioritize high-impact exposures. Control—enforce policies in real time, block, allow, or coach users. Report—generate executive-ready insights and support compliance and audits. All from a single AI Control Plane.
AI governance is now a core responsibility for CISOs. Traditional governance models are insufficient for AI. Effective governance requires visibility, automation, and control. The five-pillar framework provides a practical approach. Organizations must move from policy to enforcement.
If you're looking to build or mature your AI governance framework, AIBound is the platform security teams trust to go from shadow AI to managed AI—in under 24 hours. Visit aibound.com or book a demo to see AIBound in action.
