AI safety + compliance for telecom
Support bots that can read a customer's line records touch CPNI — data the FCC fines carriers for mishandling. EvalGuard redacts subscriber and account data before third-party LLMs see it, consent-gates outbound AI before a message is ever drafted, and preserves every interaction in a tamper-evident trail.
What ships today
Honest posture, not roadmap promises
Every checked item is in production today. In-progress items are flagged explicitly — no overclaiming, no vapor.
Built for buyer reality
Telecommunications AI use cases we ship for
Customer-support assistant with account access
The support bot answers billing and plan questions using account records — CPNI. A single response leaking another subscriber's call records or location data is an FCC-reportable failure.
EvalGuard features
- Data firewall redacts phone numbers, account IDs, and location before the LLM call
- Prompt-injection defense: 300+ attack plugins cover cross-account extraction attempts
- Per-route rate limits blunt enumeration attacks against account-lookup flows
- Tamper-evident audit log records every account-touching response
Outbound retention + upsell campaigns
AI drafts and personalizes outbound SMS + voice scripts. TCPA exposure starts at $500 per message — consent state and quiet-hours rules must gate every send.
EvalGuard features
- Consent gates at /api/v1/privacy/consent block generation for opted-out subscribers
- Output guardrails enforce approved-offer language — no invented discounts
- Brand-voice scorers keep 10,000 personalized variants on-script
- Gateway cost ledger attributes spend per campaign with daily budget caps
Network-operations copilot
NOC engineers query an assistant grounded in runbooks and topology docs. A hallucinated maintenance step on live infrastructure is an outage — answers must be faithful to the runbook, and topology data must stay in-house.
EvalGuard features
- Faithfulness scorer checks every procedure against the source runbook
- Self-hosted deploy keeps network topology inside your security boundary
- BYOK provider keys — infrastructure docs never transit shared credentials
- Tamper-evident audit log preserves every answer and its context for post-incident review
Voice-agent fraud screening
Voice AI handles account verification calls. SIM-swap fraud rings target these flows with cloned voices — identity verification needs deepfake and voice-clone detection in the loop.
EvalGuard features
- Voice-clone + deepfake guardrails on the realtime voice control plane
- Per-frame guardrail scan via the gateway realtime WebSocket proxy
- Escalation rules hand suspicious verifications to human agents
- Every verification attempt logged with signals for fraud-team review
Wire it in 60 seconds
Wrap your OpenAI client. Keep CPNI out of the prompt.
Redaction rules + consent gates + audit retention are configured once in the EvalGuard control plane. Your code only wraps the client.
import OpenAI from "openai";
import { wrapOpenAI, EvalGuardViolationError } from "@evalguard/openai";
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
projectId: "support-assistant",
metadata: { vertical: "telecom", cpni: true },
blockOnViolation: true, // refuse CPNI-leaking prompts
evalOnResponse: { failOnScore: 0.7 }, // faithfulness + brand gate
onViolation: (r) => alertPrivacyOps(r.violations),
});
try {
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: supportPrompt }],
});
} catch (err) {
if (err instanceof EvalGuardViolationError) {
// Block recorded in the audit trail. Replay via audit ID.
}
}wrapOpenAI for wrapAnthropic.Stack
Six surfaces, one platform
Eval, firewall, red-team, audit, BYOK, dashboard — every surface ships out of the box. No bolt-on vendors, no procurement cycle per capability.
Ready to ship carrier AI you can defend?
Free trial includes the data firewall, consent gates, and the tamper-evident audit trail. Self-hosted deploy available for network-adjacent workloads.
Apache-2.0 source · SOC 2 evidence engine live · full trust center