Assail launches Sidewinder, an offensive security AI that fixes its own mistakes
Autonomous security company Assail Inc. today launched Sidewinder, a rebuilt version of its Ares offensive security platform that runs on a new 31 billion-parameter model designed to audit its own findings and repair its own mistakes without human intervention.
Sidewinder is a full second-generation redesign of Ares, which Assail introduced in January, with a new interface, a new model and a rebuilt testing harness. The model was fine-tuned on 26 years of practitioner work from Alissa Knight, the company’s chief executive and chief AI officer and a longtime application programming interface hacker.
The release reframes Ares from a scanner that follows a fixed script into a system that plans its own engagements. The previous version, called Dagger, ran every target through the same sequence of phases on a single model and retained nothing once a scan ended. Sidewinder is chat-first and plan-driven, using a fleet of 12 specialized autonomous agents that work against a persistent knowledge graph of a target’s attack surface and rewrite the plan as new evidence surfaces.
The platform now uses vision-grounded analysis to drive a real browser, letting it crawl single-page applications, clear simple challenges and operate several authenticated identities at once. Assail said that approach exposes cross-account, broken object-level authorization and broken function-level authorization flaws that sit in the gap between users.
