TypeSafe opens Jev AI to public after rapid adoption

Almeida: Almeida is a TypeSafe representative who reported on Jev’s immediate adoption metrics and the company’s decision to drop its waitlist shortly after the September 15 launch. CJ Moses: CJ Moses is Amazon’s CISO and has shared examples from honeypot captures of AI agents completing multi-step attacks in under 15 minutes. Pydantic: Pydantic maintains official documentation for its Jev integration that warns about the model’s sensitivity to the ordering of Literal or Enum options and the risks of injected text steering decisions. The library explicitly states that a guard built on Jev should sit alongside deterministic checks rather than replace them, treating input state as data rather than potentially hostile content. Pydantic’s guidance aligns with TypeSafe’s own limitations documentation on prompt injection vulnerabilities. TypeSafe: TypeSafe developed Jev, a specialized model that returns structured choices, scores and probabilities rather than generating prose or code for use in agent routing and classification pipelines. The company has documented how adversarial text, including injected instructions, can influence Jev’s outputs and has positioned the model as a low-latency alternative to full LLM calls in enterprise workflows. TypeSafe opened Jev to the public on September 20 after clearing its initial waitlist and highlighted the need for complementary deterministic controls. Adam Meyers: Adam Meyers is CrowdStrike’s SVP of counter adversary operations and has discussed observed AI-driven attack campaigns that execute large numbers of commands at machine speed. Chris Goettl: Chris Goettl is Ivanti’s VP of Product Management for Endpoint Security and has described the company’s use of frontier models in vulnerability-management pipelines with a mandatory human review step for all outputs. Kayne McGladrey: Kayne McGladrey is an IEEE Senior Member and cybersecurity adviser who has tracked enterprise practices around AI agent identity and permission management. McGladrey has noted that many organizations apply human-style user profiles to agents, creating potential permission sprawl, and that frameworks such as SOC 2 and ISO 27001 have not fully addressed agent-specific identities. Risk Awareness: Both TypeSafe and Pydantic publish warnings that adversarial or injected text can steer Jev’s classification and decision outputs. Agent Evaluation: Jev is now available as a judge model in LangSmith Evals for assessing open-ended agent behavior and generating structured feedback. Security Practices: Major security vendors including Cisco, Palo Alto Networks, SentinelOne, Microsoft and CrowdStrike maintain human-in-the-loop checkpoints for high-impact AI agent actions.