ASAPP has launches Continuous Red Teaming, a new capability that integrates adversarial AI testing directly into ASAPP’s model evaluation framework. The new capability is built on Promptfoo, an AI security platform that helps enterprises detect and address vulnerabilities in AI systems during development.

Promptfoo continuously runs automated tests across ASAPP’s AI systems, screening for more than 50 vulnerability types to give enterprise customers the real-time data they need to trust their AI in production.

As AI agents shift from answering questions to taking actions, the potential for security threats increases. Static safety filters and one-time audits are no longer enough. Adversarial techniques have advanced, exploiting the same reasoning skills that make large language models effective. Enterprises using AI at scale are being asked to trust systems they can’t continuously verify. ASAPP’s Continuous Red Teaming is designed to bridge that gap.

“Enterprise AI builds trust through proven results at scale,” stated Priya Vijayarajendran, CEO of ASAPP. “Organizations using AI for essential customer interactions deserve solid proof, not just promises. Our dedication to Continuous Red Teaming guarantees we meet and exceed that commitment every single day.”

The capability operates across three critical security domains:

• Core model integrity & control: ASAPP’s agentic platform continuously tests against adversarial jailbreaking techniques, including many-shot attacks and character-level obfuscations, as well as system override attempts and probes for hallucinated authority, where a model might be coerced into making unauthorized legal or business commitments.
• Data privacy & knowledge base security: In RAG-driven deployments, retrieved data is as vulnerable as the model itself. ASAPP tests against indirect prompt injection via poisoned documents, knowledge base exfiltration, and cross-session PII leakage to ensure that sensitive data never bleeds across sessions or tenants.
• Agentic & operational security: As AI moves from conversation to action, ASAPP validates that its agents cannot be manipulated into unauthorized data access, cannot be used as proxies for internal network reconnaissance, and cannot expose underlying system architecture through tool-calling exploitation.

The core capability features a risk benchmarking framework where ASAPP tracks the Attack Success Rate (ASR) for each model update. It uses automated graders to minimize human bias in safety evaluations and aligns results with the OWASP Top 10 for LLMs and the NIST AI Risk Management Framework, ensuring audit-ready safety metrics from day one.

“CISOs and procurement teams looking at enterprise AI need more than a vendor’s word that their systems are safe,” said Khash Kiani, CISO of ASAPP. “Trust has to be earned through specific metrics, a clear approach, and a regular reporting system to share with their boards. We have created a safety architecture that automatically identifies and fixes weaknesses in real time. This is the standard that enterprise AI should aim for, and we are committed to leading the way.”