/
← Voltar ao feed
medium smb

ESET: AI accelerates attacks on SMBs as 3,000+ malicious AI agent skills found and readiness lags

1 fonte(s) · 1 atualização(ões) · publicado 16/09/2026 06:00 · coletado há 4 dias

This ESET blog post argues that AI is speeding up the threat landscape for SMBs, both by empowering threat actors and by widening the attack surface as businesses adopt AI faster than they govern it. The ESET SMB Cyber Readiness Index 2026 reports that 73% of SMBs are integrating AI, 40% lack a proper AI policy, and 45% suffered a cybersecurity incident last year. ESET analyzed 900,000 AI agent skills across popular repositories between March and May 2026 and found over 25,000 suspicious and more than 3,000 tagged as malicious. The article also cites PromptSpy, described as the first AI-powered Android spyware, which abuses Google's Gemini at runtime for persistence. The remainder is vendor positioning for managed, AI-assisted detection and response services.

Detalhes técnicos

  • Malicious AI skills, which work like browser plugins for AI agents, exfiltrate data, abuse permissions, download and execute malware, manipulate sensitive systems, override instructions through prompt injection and alter agent behavior.
  • Prompt injection, which OWASP recently branded the most dangerous LLM threat, can be delivered directly or hidden in content the AI later retrieves.
  • Other vectors include malicious links served through chatbots (phishing and malware installs) and poisoned download sources or components that AI agents interact with.
  • Attacker use of AI cited from UK government guidance: automated victim reconnaissance, faster vulnerability research and exploit development (a collapsed exploitation window), convincing localized social engineering, basic malware development, and processing of exfiltrated data.
  • SMB pain points: tool sprawl, alert fatigue, no 24/7 monitoring, longer dwell time, misconfigured security products and AI governance gaps such as shadow AI.

Remediação

  • Adopt and enforce an AI usage policy, and limit the permissions and data access granted to AI agents.
  • Vet AI skills and components before use, and protect AI conversations, agents and outputs.
  • Maintain documented, regularly reviewed incident response plans.
  • Add 24/7 monitoring, for example through a managed service, with human oversight for contextual decisions.
  • Plan for operational resilience so that minimum viable operations continue during incidents.

Linha do tempo

  1. welivesecurity.com ↗ · pub. 16/09/2026 06:00 · coletado 19/10 18:00 primeiro registro

    ESET: AI accelerates attacks on SMBs as 3,000+ malicious AI agent skills found and readiness lags

    ler a análise desta fonte

    This ESET blog post argues that AI is speeding up the threat landscape for SMBs, both by empowering threat actors and by widening the attack surface as businesses adopt AI faster than they govern it. The ESET SMB Cyber Readiness Index 2026 reports that 73% of SMBs are integrating AI, 40% lack a proper AI policy, and 45% suffered a cybersecurity incident last year. ESET analyzed 900,000 AI agent skills across popular repositories between March and May 2026 and found over 25,000 suspicious and more than 3,000 tagged as malicious. The article also cites PromptSpy, described as the first AI-powered Android spyware, which abuses Google's Gemini at runtime for persistence. The remainder is vendor positioning for managed, AI-assisted detection and response services.

    Technical Details

    • Malicious AI skills, which work like browser plugins for AI agents, exfiltrate data, abuse permissions, download and execute malware, manipulate sensitive systems, override instructions through prompt injection and alter agent behavior.
    • Prompt injection, which OWASP recently branded the most dangerous LLM threat, can be delivered directly or hidden in content the AI later retrieves.
    • Other vectors include malicious links served through chatbots (phishing and malware installs) and poisoned download sources or components that AI agents interact with.
    • Attacker use of AI cited from UK government guidance: automated victim reconnaissance, faster vulnerability research and exploit development (a collapsed exploitation window), convincing localized social engineering, basic malware development, and processing of exfiltrated data.
    • SMB pain points: tool sprawl, alert fatigue, no 24/7 monitoring, longer dwell time, misconfigured security products and AI governance gaps such as shadow AI.

    Remediation

    • Adopt and enforce an AI usage policy, and limit the permissions and data access granted to AI agents.
    • Vet AI skills and components before use, and protect AI conversations, agents and outputs.
    • Maintain documented, regularly reviewed incident response plans.
    • Add 24/7 monitoring, for example through a managed service, with human oversight for contextual decisions.
    • Plan for operational resilience so that minimum viable operations continue during incidents.