Kwal — agentic threat model

Not certain from the listing — The specific LLMs and text-to-speech/speech-to-text models are not disclosed. However, the voice-based interface introduces risks of adversarial audio injection, voice cloning exploitation, and prompt injection embedded in candidate speech to manipulate the screening outcome.

Kwal processes highly sensitive candidate PII, resume data, and emotional analysis metrics. The primary threats include data exfiltration of candidate profiles and potential data poisoning if malicious resumes or candidate inputs are ingested into its 'perfect recall' memory system.

The agent uses adaptive questioning and integrates with 45+ ATS platforms. Insecure tool integration could allow an attacker to exploit the ATS API connection, leading to unauthorized data modification or extraction from the recruitment database.

Not certain from the listing — The infrastructure supporting up to 1 million concurrent voice connections is not detailed. Potential threats include telephony/SIP-based denial of service, interception of live audio streams, and unauthorized access to call recordings.

Kwal features call monitoring and comprehensive analytics, including emotional analysis. A key threat is the potential for bias or blind spots in the emotional analysis algorithms, as well as the risk of candidates gaming the evaluation metrics.

Not certain from the listing — There is no mention of compliance certifications (e.g., SOC2, GDPR, CCPA) or bias-mitigation audits. Given its role in recruitment, it represents a high-risk AI system under frameworks like the EU AI Act, requiring strict compliance controls.

Kwal operates within a broader ecosystem by connecting directly to 45+ external Applicant Tracking Systems. A compromise of Kwal could lead to cascading security failures across these connected HR platforms, allowing lateral movement or bulk data harvesting.