At Black Hat USA 2026, Novee found GitHub workflow flaws in Claude Code, Gemini CLI and Codex that enabled RCE, credential theft and agent control in pipelines.

Learn how growing teams make AI tools reliable with clear workflows, shared rules, secure systems, and repeatable processes that improve quality and speed daily

The UK’s National Cyber Security Centre (NCSC) has warned organizations to take a measured approach toward adopting agentic AI, highlighting the growing cyber and operational risks associated with highly autonomous AI systems. In a new guidance document co-authored with international partners, the NCSC said businesses should avoid rushing into large-scale deployments of agentic AI tools without understanding the security implications. The guidance recommends starting with low-risk use cases, limiting system privileges, and maintaining strong human oversight throughout deployment. The advisory comes as organizations increasingly experiment with AI systems capable of making decisions, accessing tools, and carrying out actions with limited human involvement.

What Is Agentic AI?

Unlike traditional generative AI systems that primarily create text, images, or predictions, agentic AI systems are designed to independently pursue goals. These systems can access data sources, remember context, make decisions, interact with software tools, and even create sub-agents to complete tasks. According to the NCSC, this added autonomy is what makes agentic AI useful for areas such as cyber defense, workflow automation, and operational efficiency. However, it also introduces a wider attack surface and increases the difficulty of monitoring system behavior. The agency noted that many security risks linked to AI are not entirely new. Concerns around access control, supply chain security, monitoring, and incident response already exist in traditional IT systems. Agentic AI systems also inherit existing large language model risks, including prompt injection and jailbreaking attacks. However, the NCSC warned that the autonomy of agentic AI systems could amplify these issues, especially if organizations deploy them without proper safeguards.

Why Agentic AI Raises Security Risks

The guidance outlines several risks tied to agentic AI deployments. One of the main concerns is broader access to systems and sensitive data. AI agents may interact with external tools, APIs, or databases in ways that traditional AI applications do not. The NCSC also highlighted the possibility of unpredictable behavior. Since AI agents interpret goals autonomously, they may take actions that differ from human expectations or exceed their intended scope. Another challenge involves visibility and oversight. Autonomous systems can operate at speeds that make meaningful human review difficult, particularly in enterprise environments where multiple systems and workflows are interconnected. The guidance further noted that explaining the behavior of agentic AI systems can be more difficult than understanding conventional AI models. The combination of decision-making, tool usage, and autonomous actions creates additional complexity during incident investigations or compliance reviews.

NCSC Calls for Incremental Agentic AI Deployment

To reduce risks, the NCSC urged organizations to adopt agentic AI gradually instead of deploying it across critical systems from the outset. The guidance recommends tightly controlled pilot deployments focused on clearly defined, low-risk tasks. Organizations are also encouraged to assess whether AI is genuinely necessary before integrating autonomous agents into existing workflows. “If you cannot understand, monitor or contain an agent’s actions, it is not ready for deployment,” the guidance stated. The agency stressed that organizations should never grant unrestricted access to sensitive data or critical infrastructure. Maintaining visibility into AI system behavior and preserving meaningful human control were identified as key requirements for safe deployment.

Human Accountability Remains Essential

Despite the growing capabilities of autonomous AI systems, the NCSC emphasized that humans remain fully accountable for how these technologies are used. The guidance states that organizations should clearly define who is responsible for approving AI access, monitoring system behavior, reviewing incidents, and shutting systems down when necessary. Security teams were also advised to integrate agentic AI risk management into existing cybersecurity and governance frameworks instead of treating AI security as a separate process. Recommended practices include applying least-privilege access controls, limiting system scope, avoiding long-lived credentials, monitoring unusual behavior, and planning for incidents involving AI misuse or loss of control.

Path Forward

While warning about the risks, the NCSC acknowledged that agentic AI could deliver significant operational benefits, particularly for repetitive and low-risk tasks. The agency said organizations should focus on responsible and scalable adoption strategies built around existing cybersecurity practices and strong governance controls. The guidance ultimately encourages businesses to move carefully, test systems incrementally, and prepare for potential failures before expanding the role of autonomous AI systems across enterprise environments.

By Niall Browne, CEO and Founder, AIBound

Shadow AI is accelerating alongside artificial intelligence (AI) adoption at a pace that has outgrown most enterprise governance models. Artificial intelligence (AI) adoption is accelerating at a pace that has outgrown most enterprise governance models. According to the World Economic Forum, 87% of organizations report that AI-related vulnerabilities are now the fastest-growing cyber risk. Part of this surfaces with  employees increasingly deploying autonomous AI agents that connect to MCP servers and external AI that security teams have never assessed, quietly piping sensitive corporate data into systems no one in IT has ever audited — and no one in the C-suite knows exist. This increase in Shadow AI is creating systemic enterprise risk that can lead to unforeseen costs. Compliance frameworks like the Artificial Intelligence Act of the European Union (EU AI Act) take full effect this year introducing penalties up to 7% of global annual revenue for unmanaged AI. As regulatory frameworks begin to align with the realities of increased AI adoption, enterprises need to account for decentralized AI usage that operates outside traditional controls. This requires software that allows greater visibility, organization, and control into how AI is used and tracked across environments.

Shadow AI Is Creating a New Enterprise Attack Surface

The traditional security stack was built for a world that no longer exists — one with known assets, centralized systems, and software that asked permission before it ran. As new tools are introduced independently, usage levels evolve quickly without system checks or visibility into how these tools interact with sensitive data. Research indicates that 75% of CISOs have discovered unsanctioned GenAI tools in their environments, and only 5% feel confident they could contain compromised AI agents. Because of how easy these platforms are to access and require little onboarding, adoption is happening across teams at a rapid rate without IT involvement. Other security issues lie with employees integrating workflows with personal AI agents. These deployments allow sensitive information to be leaked or directly inputted into agents without security knowledge. Without a system in place for organizations to continuously track and evaluate how AI is being used across their enterprise systems, CISOs are left without visibility of their attack surfaces. The result is a slow-motion breach: data leaking, compliance crumbling, and governance reduced to a slide deck nobody enforces. Recurring data leaks and breaches via AI reveal the need for solutions that address this gap. Popular AI agents like ChatGPT for example, revealed a ‘ShadowLeak’ vulnerability that allowed sensitive email data to be breached through a zero-click attack. Other short lived features that rolled out last year allowed conversation sharing, leaving employee info, internal corporate strategies, and other sensitive data to be shared and indexed by search engines. Although this option only was available for a day, it was estimated that over 100,000 private chats were affected and able to be viewed with a simple search, allowing any sensitive information inputted to be publicly accessible. Other recent breaches include a Microsoft 365 Copilot bug allowing AI assistants to summarize emails labeled confidential, bypassing data loss prevention policies set up by organizations. Microsoft confirmed that a code issue allowed confidential emails data to be accessed despite organizational securities put in place. These agents are live and operational with local access to files, systems, commands, and APIs capable of executing tasks and retrieving data without clear oversight control. As AI usage continues to expand at accelerating rates, organizations need a way to better understand how these tools are used across their environments. No CISO has ever defended a perimeter they couldn't see. Shadow AI is the new perimeter — and most security teams are flying blind. Without a comprehensive inventory and control of AI usage, security teams are unable to accurately assess risks and enforce policy to maintain compliance.

Shadow AI Demands Continuous Visibility and Independent AI Control Planes

This is where adoption of independent AI Control Planes becomes vital. Independent AI Control Planes provides a way to continuously identify and assess AI activity giving security teams the visibility needed to manage emerging risks. It enables organization and categorization of AI usage across enterprises without relying on the manual entry and tracking that existing platforms demand — work no security team in a fast-moving environment can realistically keep up with. It’s undeniable: Shadow AI is not a future problem — it is already running inside your enterprise, on assets you don't own, through agents you never approved, touching data you are responsible for protecting. Every day without continuous, autonomous AI discovery is a day your attack surface grows faster than your governance can chase it. Regulators won't wait. Attackers already aren't. The CISOs who win the next 24 months will be the ones who stop pretending policy equals control and start operating on a simple truth: if you can't see it, you can't secure it — and right now, most of AI is invisible.

Disclaimer: The views and opinions expressed in this guest article are solely those of the author and do not necessarily reflect the official policy or position of The Cyber Express. The information shared is intended for industry discussion and awareness purposes only.

Organisations worldwide are being urged to prepare for a vulnerability patch wave, as security experts warn that advances in artificial intelligence (AI) could rapidly expose long-standing weaknesses across software systems. The warning comes from National Cyber Security Centre (NCSC), which says businesses must act now to strengthen their environments before a surge of critical updates arrives. In a blog, Chief Technology Officer Ollie Whitehouse highlighted that years of accumulated technical debt are now becoming a major cybersecurity risk. Technical debt refers to unresolved flaws and compromises in software that arise when organisations prioritise speed or short-term delivery over long-term resilience. According to Whitehouse, artificial intelligence is accelerating the problem. Skilled attackers are increasingly able to use AI tools to identify and exploit vulnerabilities at scale, forcing what the NCSC describes as a “correction” across the technology ecosystem. This is expected to trigger a vulnerability patch wave, with a high volume of security updates affecting open source, commercial, proprietary, and software-as-a-service platforms.

Prioritising External Attack Surfaces

As part of preparing for the vulnerability patch wave, the NCSC advises organisations to first focus on their external attack surfaces. Internet-facing systems, cloud services, and exposed infrastructure present the highest risk when new vulnerabilities are disclosed. The guidance recommends a perimeter-first approach. Organisations should secure outward-facing technologies before moving deeper into internal systems. This reduces the likelihood that attackers can exploit newly discovered weaknesses during the vulnerability patch wave. Where resources are limited, priority should be given to patching systems that are directly exposed to the internet. Critical security infrastructure should follow next. However, the NCSC cautions that patching alone will not solve every issue. Legacy and end-of-life systems remain a major concern. Many of these technologies no longer receive security updates, leaving organisations vulnerable even during a vulnerability patch wave. In such cases, businesses may need to replace outdated systems or bring them back into supported environments, especially if they are externally accessible.

Preparing for Faster and Large-scale Patching

The expected vulnerability patch wave will require organisations to rethink how they manage updates. The NCSC is urging businesses to prepare for faster, more frequent, and large-scale deployment of security patches, including across supply chains. Several key measures have been recommended:

• Enable automatic updates wherever possible to reduce operational burden
• Adopt secure “hot patching” to apply fixes without service disruption
• Ensure internal processes support rapid and large-scale updates
• Use risk-based prioritisation models such as Stakeholder Specific Vulnerability Categorisation (SSVC)

Whitehouse noted that organisations must be ready to accelerate patching timelines when critical vulnerabilities are actively exploited, particularly those affecting internet-facing systems. At the core of this approach is an “update by default” policy. This means applying software updates as quickly as possible, ideally through automated processes. While this may not always be feasible for safety-critical or operational technology systems, the NCSC says it should form the foundation of modern vulnerability management strategies.

Beyond Vulnerability Patch Wave: Addressing Systemic Risks

The NCSC emphasises that the vulnerability patch wave is only part of a broader cybersecurity challenge. Patching addresses immediate risks, but it does not eliminate the underlying causes of technical debt. Technology vendors are being encouraged to build more secure systems from the outset. This includes adopting memory safety and containment technologies such as CHERI, which can reduce the likelihood of exploitable vulnerabilities. For organisations operating critical services, strengthening cybersecurity fundamentals is equally important. Frameworks such as Cyber Essentials and sector-specific resilience models can help reduce the impact of breaches and improve overall security posture. Additional guidance has also been issued for high-risk environments, covering areas such as privileged access workstations, cross-domain security architecture, and threat detection through observability and proactive hunting.

Organisations Urged to Act Now

The NCSC has made it clear that preparation cannot be delayed. The anticipated vulnerability patch wave is expected to impact organisations of all sizes and sectors. Businesses are advised to review their vulnerability management processes, assess their exposure, and ensure their supply chains are also ready to respond. Larger organisations, in particular, are encouraged to seek assurance from both commercial and open-source partners. As Whitehouse concluded, readiness for the vulnerability patch wave will depend on proactive planning, strong fundamentals, and the ability to respond quickly at scale.

Remove unwanted objects from video effortlessly with AI in 2026. Learn step-by-step methods, best tools, and pro tips to clean up your footage like a professional.

On February 20, 2026, AI company Anthropic released a new code security tool called Claude Code Security. This release coincided with the highly sensitive period of global capital markets to AI technology subverting the traditional software industry, which quickly triggered violent fluctuations in the capital market and caused the fall of stock prices of major […]

The post Insights into Claude Code Security: A New Pattern of Intelligent Attack and Defense appeared first on NSFOCUS, Inc., a global network and cyber security leader, protects enterprises and carriers from advanced cyber attacks..

The post Insights into Claude Code Security: A New Pattern of Intelligent Attack and Defense appeared first on Security Boulevard.