Hackaiz Academy · AI and agents

AI agent security (OWASP Agentic)

Risks of agents using tools, MCP and Skills.

Automated technical check — neither certification nor legal advice.A service operated by HACKTUALIZ Inc.
  • Least-privilege tools
  • Human approval for sensitive actions
  • Component provenance
  • Static import gate for Skills, MCP servers and packages
  • Unknown components are never executed outside a sandbox

No. Analysis is static.

https://genai.owasp.org/

Control points

6 automatic · 0 assisted (your answers or documents) · 1 to be proven by a person

  • Goal hijacking and memory poisoningAutomatic
  • Tool misuse and command executionAutomatic
  • Exposed or unencrypted MCP serversAutomatic
  • Agent output executed uncheckedAutomatic
  • Runaway loops and unbounded consumptionAutomatic
  • Action traceability without secretsAutomatic
  • Human approval of sensitive actionsHuman

Check your project: domain ownership proof and consent required, each check is billed in credits by actual work. Results are shown as verified points with evidence, never as a certification.