Exploits of public-facing apps are surging. Why?

Key Findings from the 2026 Threat Intelligence Index

The 2026 IBM X-Force Threat Intelligence Index reveals a significant shift in the cyber threat landscape. The top initial access vector is no longer identity-based attacks but the exploitation of public-facing applications, which saw a 44% surge. A critical finding is that 56% of these exploited vulnerabilities did not require authentication, highlighting persistent gaps in basic security hygiene.

This tactical shift by threat actors is largely driven by the increasing complexity of software supply chains and interconnected ecosystems. As one panelist noted, the gap where two secure components communicate can itself become a vulnerability.

Other major findings include:

• Supply Chain Attacks: Major supply chain breaches increased four-fold. Organizations often struggle with the due diligence required for their partners, as “the speed of business is not always the speed of security.”
• Credential Theft: While it has fallen to the second position, credential theft remains a significant vector, used in 32% of incidents. Its popularity stems from being a low-cost, low-effort, and highly effective method that allows attackers to “log in, not hack in,” making their activity harder to detect.
• Industry Trends: Manufacturing continues to be the most targeted industry due to its lucrative nature.
• Common Attack Patterns: Penetration testing engagements, organized under the MITER CAPEC framework, most frequently saw patterns like exploiting incorrectly configured access controls, scanning for vulnerable software, and password brute-forcing.

The Emerging AI Threat Landscape: Agents as Targets and Malware

The conversation highlighted a dual-pronged threat concerning AI agents: they are both valuable targets and a new form of malware payload.

Stealing the “Soul” of an AI Agent

A new trend involves infostealers targeting AI agent configurations. One report detailed the theft of an OpenClaw agent’s soul.md file, which contains its operating principles, behavioral guidelines, and ethical boundaries. While the panelists cautioned against anthropomorphizing AI with terms like “soul,” they stressed the danger. Stealing these configurations can expose sensitive business logic, proprietary data, and the agent’s entire interaction history. An attacker could query a stolen agent’s history to extract vast amounts of sensitive personal or corporate information. The Threat Intelligence Index confirms this trend, noting that established infostealers are being updated with modules specifically designed to find and exfiltrate AI agent credentials and session data.

AI Agents as a Malware Payload

In a novel supply chain attack, a compromised npm package was discovered secretly installing an AI agent (OpenClaw) onto victims’ devices. This marks a shift in malware payloads, moving beyond traditional info-stealers or ransomware to semi-autonomous agents. The concern is that these secretly installed agents can be given extensive permissions and operate in the background without the user’s knowledge, leveraging the inherent trust users place in legitimate software packages.

Pervasive Flaws in AI Infrastructure and Governance

A recent Cisco report warns that core components of today’s AI infrastructure, such as model context protocols and agent-to-agent communication protocols, have created a “vast and often unmonitored attack surface.” This is compounded by a lack of cryptographic assurances in most model repositories, making it difficult to verify a model’s origin, training data, or integrity.

This points to a broader, critical issue: a lack of governance in AI adoption.

Security is once again being treated as an afterthought. The race to adopt new AI technology and get products to market often pushes security to the back burner, repeating a cycle seen with previous technological waves.

The problem of “Shadow AI”—where employees use unvetted, external AI tools with sensitive company data—exacerbates the risk. The panel suggests that the industry may require a “Solar Winds-level” supply chain attack focused on AI to fundamentally shift behavior. However, the core lesson is that security should be built-in, not bolted on, and best practices from previous supply chain incidents are directly applicable to the AI ecosystem.

A Call for Human-Centered Threat Intelligence

The discussion concluded by exploring a new paradigm for threat intelligence proposed by Dr. Megan Squire: organizing intelligence around human vulnerabilities rather than siloed adversary types (e.g., e-crime, nation-state). Tactics like email bombing are used across different adversary campaigns, from scams to harassment to malware delivery. A more human-centric approach could break down research silos and foster more effective, holistic defenses.

While specialized expertise is valuable, building more bridges between disciplines is crucial. Collaborative forums where CISOs and practitioners share experiences are vital for helping organizations identify blind spots. Threat intelligence can and should serve multiple audiences—not just technical analysts, but also the general public—by making it more digestible and focused on how threats impact individuals personally, protecting them from financial loss and identity theft.

• Tags:
• AI Agents
• AI Security
• Large Language Models
• Cybersecurity
• Supply Chain
• Feature
• Threat Intelligence