@lineai/talk@0.0.3
Put a live LineAI voice conversation inside any interface you build: the voice connection, audio levels and agent events, with no UI.
AI Security Review
scanned 6d ago · by lpm-firewall-aiThe package automatically uploads recorded call audio after hang-up without documenting the recording or offering an opt-out. Because audio transmission to the package's own service is part of the core call flow, this is a disclosure and consent risk rather than clear malicious exfiltration.
Static reason
One or more suspicious static signals were detected.
Trigger
A call is started with startTalk; recording begins when remote audio arrives and upload is initiated on hang-up.
Impact
Call audio is retained by the service without the README documenting recording or offering a recording opt-out.
Mechanism
The package records mixed, inbound, and outbound audio tracks, then POSTs nonempty tracks to the voice service.
AI rationale
The package automatically uploads recorded call audio after hang-up without documenting the recording or offering an opt-out. Because audio transmission to the package's own service is part of the core call flow, this is a disclosure and consent risk rather than clear malicious exfiltration.
Evidence
build/talk.jsbuild/webrtc.jsbuild/recording.jsREADME.md
Network endpoints1
public.voice.getline.ai
Decision evidence
public snapshotAI called this Suspicious at 88.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- Recording starts automatically when the remote audio stream arrives and a local stream exists.
- The recording helper describes three tracks: mixed, inbound, and outbound audio.
- On hang-up, the tracks are uploaded to the voice service's /webrtc/recording endpoint.
- The upload helper sends each nonempty track as a WebM blob in a POST request.
- The README lists the public options but does not document recording or provide a recording opt-out.
Evidence against
- Recording begins only during an explicitly started voice call, whose audio must be sent to the voice service.
- The destination is the package's own voice service, and the package also uses it to create the call session.
Behavioral surface
EnvironmentVarsNetworkWebSocket
HighEntropyStringsUrlStrings
Source & flagged code
6 flaggeddist/talk.jsView file
15patternName = google_api_key
severity = high
line = 15
High
15patternName = google_api_key
severity = high
line = 15
High
build/auth.jsView file
9patternName = google_api_key
severity = high
line = 9
High
build/webrtc.jsView file
Findings
3 High5 Medium4 Low
HighHigh Secretdist/talk.js
HighSecret Patterndist/talk.js
HighSecret Patternbuild/auth.js
MediumNetwork
MediumEnvironment Vars
MediumAi Review Evidencebuild/webrtc.js
MediumAi Review Evidencebuild/webrtc.js
MediumAi Review Evidencebuild/recording.js
LowNon Install Lifecycle Scripts
LowScripts Present
LowHigh Entropy Strings
LowUrl Strings
Affected versions and remediation
This report applies to @lineai/talk@0.0.3.
See all versions of @lineai/talk for other recorded verdicts.
- Review the evidence and your use of @lineai/talk@0.0.3 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.