1
Guardrails 1 company
Guardrails sit between the user and the model. They check every message going in, and block one that tries to trick the model into ignoring its instructions, called prompt injection. They also check every answer coming out before the user sees it.
2
Hiding sensitive data from the model 1 company
Sensitive data, such as account numbers, is removed or replaced with stand-in values before a model or an AI agent sees it.
ProtectoProtecto removes or replaces sensitive data, such as a bank's account numbers, before an AI model or agent sees it.See alsoData security and privacyHow it all worksGenerative AI applicationsTurning a language model into a Hindi support assistant.
See how it works →What's changing
AI agents are the new insiders.
An AI agent can log in to systems, read files and take actions, like an employee. If it is tricked, it can do the damage an insider could. So agents now need what employees have: their own identity, only the access they need, and a record of what they did.
Elsewhere in Cybersecurity:Email security and anti-phishingIdentity and access management (IAM)KYC, identity verification and fraud preventionEndpoint security and device managementNetwork security, firewalls and SASECloud and SaaS securityApplication and API securityData security and privacySIEM and security operations (SOC)Managed security services (MDR and MSSP)Vulnerability management and penetration testingPhysical security, surveillance and trust and safety
Companies: ingrid's company map. Hiring signals: postings feed of 5 Nov 2026.