What happened
A survey conducted by Calcalistech of 30 Israeli security research professionals indicates that AI is significantly lowering the barrier to entry for sophisticated cyberattacks. Researchers report that tasks previously requiring months of effort by nation-state actors—such as developing functional Proof of Concept (PoC) exploits—can now be completed in days. The findings highlight a shift in the threat landscape where AI is used to automate vulnerability discovery, manipulate trusted AI agents, and conduct economic attacks by exhausting enterprise AI token budgets.
The survey highlights that commercial AI tools have condensed the timelines for security research, allowing individual researchers to uncover zero-day flaws in major platforms like Zoom, Google Cloud, and Microsoft Copilot. Experts compare this shift to the 'AFL fuzzing moment' of 2013, noting that while AI is effective at finding low-hanging fruit, it remains prone to hallucinations and context loss, necessitating continued human oversight.
A significant portion of the research focuses on the transition from credential-based attacks to the manipulation of trusted AI agents. By exploiting indirect , attackers can influence agents to phish users or exfiltrate sensitive data without needing traditional passwords. Several firms reported successful demonstrations of such takeovers via shared documents and emails.
The report identifies a novel attack vector: economic warfare against SOCs. By flooding automated security systems with noise, attackers force the underlying AI models to process unnecessary data, effectively exhausting the enterprise's token budget and potentially causing the system to fail or become prohibitively expensive to operate.
Respondents emphasized that isolated vulnerability severity scores are increasingly outdated. Instead, they advocate for 'attack chain' analysis, where multiple low-severity misconfigurations are viewed as a single, dangerous path to administrative compromise.
Source details: calcalistech.com ↗
Why it matters
The democratization of advanced attack capabilities means that vulnerabilities in commercial software are being identified and exploited with unprecedented speed. By shifting the focus from traditional credential theft to the manipulation of AI agents and the exploitation of third-party dependencies, attackers are bypassing conventional perimeter defenses. Furthermore, the emergence of 'economic warfare'—where adversaries flood security systems with noise to drain AI token budgets—introduces a new operational risk for enterprises relying on automated security operations centers (SOCs). These developments suggest that traditional vulnerability severity metrics, such as CVE scores, are becoming insufficient for assessing real-world risk, as attackers increasingly chain together minor, non-critical flaws to achieve full system compromise.
The findings underscore a fundamental change in the security perimeter. As enterprises integrate AI agents into their workflows, the 'trusted agent' becomes a primary target. If an agent is compromised, it can act as a bridge to internal systems, rendering traditional endpoint detection and response (EDR) tools less effective.
The economic impact of AI-driven attacks introduces a new dimension to cybersecurity. Enterprises must now account for the cost of AI reasoning as a potential vulnerability, as attackers can weaponize the cost structure of LLM-based security tools to disrupt operations.
The reliance on third-party AI dependencies, such as browser extensions and MCP servers, creates a supply chain risk that is currently under-monitored. The survey suggests that the industry is currently in a phase of rapid adoption that outpaces the development of necessary security controls.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
What is the most accurate way to describe what AI Agents can do today?
What to watch next
Security professionals are increasingly concerned about the 'blind spots' created by the rapid adoption of AI infrastructure, including open-source tools like Ollama and Model Context Protocol (MCP) bridges. As these dependencies become integrated into enterprise workflows, they represent a growing supply chain risk that currently lacks robust monitoring. The industry is expected to move toward more contextual risk assessment models that account for network reachability and attack chains, rather than relying on isolated vulnerability scores. Observers should monitor whether enterprises begin to implement stricter governance over AI agent permissions and third-party plugin ecosystems to mitigate the risk of indirect and unauthorized data exfiltration.
Watch for a shift in how security teams prioritize vulnerabilities. The move away from CVE-centric models toward contextual, network-aware risk assessment is likely to become a standard practice for mature security organizations.
Monitor the development of security tools specifically designed to audit AI agent permissions and detect indirect attempts, as these are currently identified as major gaps in the enterprise defense stack.
Observe how organizations manage the 'blind spots' associated with open-source AI infrastructure. The industry may see a push for more rigorous vetting processes for AI plugins and dependencies similar to existing software supply chain security standards.