Organisations can now autonomously find and fix software vulnerabilities using Parcata, a new platform from Leidos.
The model-agnostic tool harnesses a diverse set of large language models to accelerate vulnerability detection in first and third-party software. Leidos said it can patch zero-day vulnerabilities in realtime, closing the gap between discovery and remediation that attackers exploit.
Being model-agnostic is central to the pitch. Rather than depending on a single frontier model, Parcata draws on several, which Leidos said ensures the mission is not tied to any one vendor's system. That design also lets the platform adapt as models improve or as particular models prove better suited to specific tasks.
The platform was developed by the company's Kudu Dynamics team, with technology validated through DARPA's AI Cyber Challenge, a United States programme that pitted automated systems against each other to find and fix flaws in real software. It has demonstrated its capabilities through recent internal technical exercises and is being run against Leidos code before deployment to mission environments.
"The use of AI by cyber adversaries has changed the game. We must detect and remediate before a breach," said Jason O'Connor, president of Leidos Intelligence.
O'Connor said the tool would help organisations stay ahead of evolving threats and make sense out of the chaos, ensuring the mission was not dependent on any single frontier model.