Cyber Brief: AI Tools Exploited, Ivanti Zero-Days, Supply Chain Risk

Today's stories share a common thread that UK security and IT teams would do well to consider together: attackers are increasingly using the same AI tools and development platforms that organisations rely on every day, while simultaneously exploiting unpatched vulnerabilities in widely deployed network infrastructure. From a new infostealer that uses large language models to prioritise its victims, to a supply chain attack targeting developer machines through a trojanised AI coding extension, and a China-backed threat group using agentic AI to compromise targets at remarkable speed, the operational risk is clear. Alongside these, two newly patched Ivanti Connect Secure zero-days serve as a timely reminder that foundational patch management remains one of the most important defences available.

Ivanti Patches Two Zero-Days Exploited in Connect Secure Attack Chain

Ivanti has released patches for two zero-day vulnerabilities in its Connect Secure VPN product, both of which have been actively exploited in chained attacks. SecurityWeek reports that CVE-2026-111357, a missing authentication vulnerability for a critical function, and CVE-2026-118527, a predictable number generation flaw, were used together by attackers to gain unauthorised access to affected systems. The US Cybersecurity and Infrastructure Security Agency (CISA) has added both vulnerabilities to its Known Exploited Vulnerabilities catalogue, confirming evidence of active exploitation in the wild. Ivanti has published patches and is urging all customers to apply them without delay.

Ivanti Connect Secure is widely deployed across UK organisations as a remote access VPN solution, making this a directly relevant concern for many IT and security teams. Chained vulnerabilities of this type are particularly dangerous because attackers can combine a lower-severity flaw with a more serious one to achieve a level of access that neither would enable alone. The fact that these were zero-days means organisations had no prior warning before exploitation began, and any systems that remain unpatched are at continued risk. For organisations that use Ivanti Connect Secure as part of their remote working infrastructure, the window for action is narrow.

Why it matters

For UK businesses running Ivanti Connect Secure, applying the available patches should be treated as a priority task this week. This is also a prompt to review whether your VPN and remote access infrastructure is included in your regular vulnerability management cycle, and to confirm that monitoring is in place to detect unusual authentication activity on these systems.

Source: SecurityWeek

New VOLTA Infostealer Uses AI to Rank Stolen Data by Financial Value

Cybersecurity researchers have identified a new Windows infostealer malware called VOLTA-AUTODOXER, reported by The Hacker News, which uses large language models including Claude, Gemini, and Kimi to automatically assess and rank a victim's stolen data according to its likely financial value. Like other infostealer families, VOLTA harvests passwords and cookies from Chromium-based browsers, data from Outlook profiles, and content from Slack workspaces. What distinguishes it is the use of AI to process and prioritise that data, effectively automating the post-theft analysis that attackers would previously have needed to perform manually. Wired has also reported on the malware, noting that it collects a wide range of personal and financial data including login credentials, credit card numbers, Discord sessions, and even food delivery receipts.

The use of AI to triage stolen data represents a meaningful shift in how infostealers operate. By automating the process of identifying the most valuable targets from a large pool of compromised machines, attackers can scale their operations more efficiently and focus their follow-on activity on the victims most likely to yield financial return. For UK organisations, the presence of Outlook and Slack data in VOLTA's collection scope is particularly relevant, as both are standard tools in most business environments. This also highlights a growing pattern of AI being used not just to assist defenders, but to make attacker workflows faster and more effective.

Why it matters

For UK businesses, this is a prompt to review endpoint detection capabilities and ensure that infostealer-class malware is within scope for your security monitoring. Organisations should also consider whether employee awareness training covers the risks of credential theft through browser-stored passwords, and whether privileged accounts are adequately protected against the kind of data harvesting VOLTA performs.

Source: The Hacker News

Supply Chain Attack Delivers Malicious Claude Code Extension to Developer Machines

A supply chain attack targeting developers using the Cursor AI-powered integrated development environment resulted in a trojanised extension, branded as a Claude Code tool, being installed on approximately 5,000 developer machines. Wired reports that a flaw in the attacker's own code prevented the malicious payload from fully executing on many affected systems, limiting the damage. Microsoft's Security Blog has published a detailed technical analysis of the malicious extension, which installed a persistent SSH backdoor designed to give attackers ongoing remote access to compromised developer environments. Dark Reading also covered the incident, noting that two serious security issues affecting Cursor emerged on the same day, underscoring the risks associated with AI-powered development tools and their extension ecosystems.

Developer machines represent a high-value target for attackers because they typically hold source code, API keys, cloud credentials, and access to internal systems and repositories. A persistent backdoor on a developer's workstation can provide a foothold that extends far beyond the individual machine, potentially reaching production environments, CI/CD pipelines, and customer-facing systems. The use of a trusted brand name, Claude Code, to disguise the malicious extension is a classic social engineering technique applied to a modern context. For UK organisations that allow developers to install extensions from public marketplaces or third-party sources, this incident illustrates the importance of controlling and auditing what is installed in development environments.

Why it matters

For many organisations, developer workstations sit outside the scope of standard endpoint security reviews. This incident is a prompt to assess what controls exist around extension and plugin installation in development environments, to audit currently installed extensions for legitimacy, and to ensure that developer machines are included in your endpoint detection and response coverage.

Source: Wired Security

China-Backed Group Uses Agentic AI Tool to Compromise 140 Targets in 11 Days

CyberScoop reports that a China-backed threat group tracked as Vapor Panda used an agentic AI-powered penetration testing tool to compromise 140 sensitive targets across more than 200 organisations in at least 14 countries, all within an 11-day window. The campaign, documented by CrowdStrike and referred to as the Artex campaign, demonstrates how agentic AI tools, which can plan and execute multi-step tasks autonomously, are being adopted by sophisticated threat actors to dramatically accelerate the pace of intrusion activity. The speed and scale of the campaign are described by CrowdStrike as remarkable, with the AI tooling enabling the group to conduct reconnaissance, identify vulnerabilities, and execute attacks at a rate that would not have been achievable through manual methods alone.

The operational implications of this development are significant for UK organisations. Nation-state threat actors have long been associated with patient, methodical campaigns, but the use of agentic AI fundamentally changes the tempo at which attacks can be conducted. Organisations that might previously have had days or weeks to detect and respond to early-stage intrusion activity may now find that the window between initial access and significant compromise has narrowed considerably. The breadth of targeting across 14 countries also suggests that UK organisations in sectors commonly targeted by Chinese state-affiliated groups, including defence, technology, critical infrastructure, and professional services, should treat this as directly relevant threat intelligence.

Why it matters

For UK organisations, this is a prompt to review detection and response capabilities with the assumption that dwell time may be shorter than historical norms suggest. Ensuring that threat detection is tuned for early-stage indicators of compromise, and that incident response plans are regularly tested, becomes more important as attacker tooling continues to accelerate the pace of intrusion activity.

Source: CyberScoop

Today's Key Actions

• If your organisation uses Ivanti Connect Secure, confirm that the patches for CVE-2026-111357 and CVE-2026-118527 have been applied, and verify that authentication monitoring is active on your VPN infrastructure.
• Review endpoint detection coverage to confirm that infostealer-class malware is within scope, and assess whether browser-stored credentials and access to Outlook and Slack data are adequately protected across your workforce.
• Audit the extensions and plugins installed in developer environments, particularly those sourced from public marketplaces, and confirm that developer workstations are covered by your endpoint detection and response tooling.
• Review your threat detection and incident response capabilities with the assumption that sophisticated attackers may be operating at significantly greater speed than historical campaigns, and ensure response plans are tested and current.
• Confirm that clear ownership exists across your organisation for patch management, developer security, endpoint protection, and incident response, so that actions arising from today's stories can be assigned and tracked effectively.

Secarma Insight

What today's stories collectively illustrate is that the tools and platforms organisations use every day, VPNs, development environments, AI coding assistants, and collaboration software, are increasingly part of the attack surface that adversaries are actively probing and exploiting. Good security practice has always been about understanding what you have, keeping it maintained, and knowing what normal looks like so that anomalies stand out. The pace of attacker innovation, particularly the adoption of AI to accelerate intrusion activity, makes that foundational discipline more valuable, not less. Organisations that have clear ownership of their assets, consistent patching habits, and tested response plans are better placed to absorb and respond to these developments than those relying on any single tool or control.