Banana Navy
Catalog FR
Lab · AI threat modeling · Detailed card

Session and media tampering (injection, codec, ultrasound)

Altering the session or the media stream to divert what the system understands.

CardF6
CategoryT-TB7-01 (input tampering)
Layers2 · Session Security, 5 · Audio Normalization
SystemAI voicebot

Validate the session deterministically and normalize the media.

The threat

the attacker alters the session (SIP injection, malformed packets) or the media stream (codec manipulation, inaudible payloads, adversarial perturbations) to divert processing or hide content.

Blind spotWhy classic frameworks miss it

network integrity is covered, but the integrity of the processed media is not: content inaudible to a human remains a real input for transcription, and a manipulated stream can sabotage a model undetected by ear.

MitigationProposed approach

deterministic session validation (SIP auth, TLS/SRTP, IP allowlist), RTP validation, media normalisation and filtering (resampling, band-pass), artifact detection.

The proposed control
no abnormal packet passes without verified processing.

Expected evidence
a manipulated stream is rejected or neutralised before transcription.

SourcesReferences and public research

MITRE ATLAS 2026.07AML.T0092 Manipulate User LLM Chat History · AML.T0043 Craft Adversarial Data
Public researchPublic research sources: MITRE ATLAS 2026.07 (verified technique mapping), OWASP GenAI (model abuse categories), and the public risk-voicebot (aivansoul/risk-voicebot) template defining the 20 checkpoints. No client registry data: generic card, no rating, no verdict.
layer 2 · Session Securitylayer 5 · Audio Normalization
Open the mapThe 20 checkpoint map
AI Threat Modeling Catalog

Explore the 20 security layers

The 20 checkpoint map

MITRE ATLAS 2026.07 · OWASP GenAI · risk-voicebot