newsfilter
← Latest News
ACCESS Newswire·

Innodata Releases the First Stage of Its AI Cyber Training Suite to Enable AI Coding Agents to Write - and Repair - Secure Code

INOD
  • Innodata Inc. (NASDAQ: INOD) released the first stage of its AI Cyber Training Suite on August 4, 2026, to help model builders & enterprises secure AI-generated code.
  • The suite comprises twelve datasets & evaluation systems built by hand from thousands of real-world security flaws discovered over the past ten years.
  • Data covers multiple programming languages including Python, TypeScript, JavaScript, Rust, C, & Go, across platforms like Linux, macOS, Android, Windows, AWS, & GCP.
  • Each security flaw was manually reconstructed in sealed offline environments to rigorously test an AI's ability to patch vulnerabilities while verifying the fix prevents the original attack & maintains functionality.
  • Testing of ten leading AI models showed open-weight models repaired at most 23% of verified flaws, but fine-tuning on the suite's data increased patching success from 18.4% to 41.2% in a single round.
  • The tool enables enterprises to modernize legacy systems & add features using AI agents while minimizing the risk of reintroducing security vulnerabilities.
  • Innodata CEO Jack Abuhoff stated the suite allows businesses to confidently deploy agent-generated code into production without requiring manual review by professional security teams for every line.
  • Yonatan Brander, lead of Innodata's cybersecurity data engineering team, noted the data provides confidence that AI agents will not introduce new holes during refactoring or modernization.
  • The AI Cyber Training Suite is immediately available to model builders & enterprises to train coding agents, harnesses, & tools to produce secure code.
  • Innodata is a global data engineering company with a 36+ year legacy focused on providing trusted data, evaluation frameworks, & human expertise for AI systems.