TypeSafe’s Jev ‘even more susceptible to attack’ than LLMs, says security vendor

Summary

HiddenLayer researchers reported that TypeSafe's Jev, a streamlined model TypeSafe classifies as a 'System One' model, is more vulnerable than traditional LLMs to some attacks, including adversarial text attacks and prompt injection. Their tests showed that the TokenBreak attack, which normally fails against larger LLMs, works against Jev out of the box, and that a manipulated advertisement caused a browser agent to click away from a user's purchase on a simulated checkout page. TypeSafe did not respond to requests for comment.

Related items

• HighCVE-2026-76066: IntentKit authorization bypass lets callers execute owner-restricted toolsSimilar attack · NVD/CVE Database
• LowJev as a judge: How counterfeit policy steers JevSimilar attack · Datadog Security Labs
• InfoSafeDriveVLA: Navigation-Conditioned World Model Dreaming for Conflict-Aware End-to-End Autonomous DrivingSimilar attack · Arxiv (cs.RO + cs.CV security)
• LowAgents at the Edge: Extending AI Security Governance to Physical AI with DefenseClawSimilar attack · Cisco AI Blog
• InfoDiagnose Before You Refine: Multi-Dimensional Judge Feedback for Query-Efficient Jailbreak AttacksSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)