Cybersecurity-Insiders AI Agents Expand Enterprise Attack Surface as Most Organizations Report Malicious Agent Incidents
Article Content
- •IRONSCALES reports 77% of surveyed organizations had an AI agent act on malicious, misleading or unauthorized content in the past 12 months.
- •About 80% of surveyed organizations give AI agents access to internal systems, including finance and payment systems at 30%.
- •Inboxes were reported as the most common entry point, with malicious email or messages acting on AI tools in 42% of incidents.
- •Experts predict AI workstations will move to employee desks, requiring IT teams to manage Linux-based devices and AI-specific visibility gaps.
- •The sources argue SIEM platforms will evolve rather than disappear, with AI agents needing historical context to investigate accurately.
Cybersecurity-Insiders' 2027 outlook piece says organizations are beginning to integrate autonomous agents into security operations, software development and IT management, and that AI hardware is moving from data centers to employee desks. Gravwell and Fleet Device Management executives quoted in the article argue SIEM platforms will evolve rather than disappear, and that IT teams will inherit management of AI workstations running Linux and locally hosted models. A second article, citing IRONSCALES' third annual Cybersecurity Threat Report, reports that 96% of respondents use AI agents that can act on employees' behalf, with 52% using them widely. The report states only 19% limit agents to the public web, meaning roughly four in five organizations give agents access to internal systems. The most common systems reached by agents are ticketing and helpdesk (67%), cloud storage (49%), email and messaging (49%), internal file repositories (49%), and finance, billing and payment systems (30%). The report states that 77% of respondents said an AI agent or tool acted on instructions or content that turned out to be malicious, misleading or unauthorized in the past 12 months. Among organizations with incidents, the two most common types were an AI tool acting on a malicious email or message (42%) and an attack on the agent itself, with the remaining detail cut off in the source. The articles do not name specific CVEs, patches or IOCs.
Ask AI about this cluster
Answers cite the sources they use
Timeline
More articles in this cluster (2)
Following this threat?
Track Ironscales in your own feed — you're alerted when they show up in new reporting, leak sites or exploitation.
Free account · no card needed
Common questions
Do most organizations give AI agents access to internal systems?
Which systems are most exposed to AI agents?
What were the most common agent incident types?
Continue Reading
Lazarus Group Targets 30 South Korean Servers Using PHP Vulnerabilities North Korea-linked Lazarus Group has targeted 30 servers in South Korea as part of a larger cyber operation involving 581 servers across 10 countries. The attacks exploit two critical remote code execution vulnerabilities in PHP (CVE-2026-109618 and CVE-2026-109619), allowing unauthorized server command execution. The…