Most conversations about AI voice agents focus on the model — how natural it sounds, how well it handles interruptions, how accurately it recognizes intent. Very few focus on what happens at the carrier layer when that AI agent makes an outbound call. That is where a significant and underappreciated failure mode lives.
STIR/SHAKEN attestation determines how the receiving carrier evaluates your call before it reaches a consumer. For AI voice operations running at scale, getting attestation wrong means calls that do not connect — regardless of how good the AI is.
Why Outbound AI Calls Face Specific Attestation Challenges
STIR/SHAKEN is a call authentication framework that assigns a trust level — A, B, or C — to every outbound call based on whether the originating carrier can verify that the calling party is authorized to use the number they are calling from.
A-level attestation is the strongest signal. It tells the receiving carrier: we know who is calling, we have verified they own this number, and we have confirmed both. Calls with A-level attestation are treated with the least suspicion by downstream carriers and analytics engines.
AI voice agents introduce specific challenges that can degrade attestation below A-level if the infrastructure is not set up correctly.
The first is number provisioning. AI voice deployments frequently use numbers that are provisioned through an AI platform's bundled PSTN layer — Twilio Programmable Voice being the most common. When numbers are provisioned inside a third-party platform rather than directly through a carrier relationship, the originating carrier may not be able to confirm the calling party's authorization for that specific number. The result is B-level attestation, even when the organization legitimately owns the numbers.
The second is routing complexity. AI voice calls often traverse multiple SIP hops before reaching the PSTN — from the AI platform, through an API layer, through a PSTN gateway, to the carrier. Each hop that does not properly pass the STIR/SHAKEN PASSporT token degrades the attestation level at the destination. A call that starts with A-level attestation at the originating carrier can arrive at the terminating carrier with C-level attestation because of a gap in the chain.
The third is scale. AI voice agents can generate high call volumes quickly. Carriers and analytics engines pay attention to volume patterns. A new number pool generating large call volumes without established attestation history is evaluated with more scrutiny than an established number pool with consistent A-level attestation signals.
What Degraded Attestation Does to Contact Rates
The practical effect of B or C-level attestation is higher termination rates and elevated Spam Likely labeling risk.
Carriers — particularly the major wireless networks — factor attestation level into their call treatment decisions. A call with C-level attestation from an unrecognized number, generating call patterns consistent with automated dialing, is a high-risk profile. The carrier may display a warning label, reduce the ring duration, or terminate the call entirely before it reaches the consumer.
For an AI voice operation running outbound appointment reminders, collections follow-ups, or customer service callbacks, this is a direct contact rate problem. The AI model performs perfectly. The call never reaches the consumer. The attestation signal is what killed it.
This failure mode is silent in most deployments because AI platforms report on calls that connected, not on the attestation level of calls that did not. Without carrier-layer instrumentation, the degradation is invisible until contact rates have already fallen enough to trigger investigation.
How Proper Attestation Is Maintained for AI Voice
The foundation is number provisioning through a direct carrier relationship. Numbers provisioned directly in a carrier network — rather than inside an AI platform's PSTN layer — carry the carrier's full attestation capability. The originating carrier can confirm both the customer identity and the number authorization, which is the requirement for A-level attestation.
The SIP routing chain matters equally. Every hop between the AI platform and the PSTN must properly handle the PASSporT token. This requires correct SBC configuration and carrier coordination — not something that happens automatically when an AI platform is pointed at a PSTN gateway.
Ongoing monitoring is the operational layer. Attestation rates should be monitored by carrier path and by number pool continuously, not checked once at deployment. Degradation in attestation level is an early warning signal for contact rate problems. Catching it before it affects consumer-facing call outcomes requires visibility at the carrier layer.
Volume management is the fourth element. New number pools should be ramped carefully — not because of any regulatory requirement, but because carrier analytics engines build trust signals over time. A number pool with consistent A-level attestation and appropriate call patterns accumulates positive signal. A new pool dialing at high volume from day one has no signal history to offset the volume flag.
The Bottom Line
STIR/SHAKEN attestation is not an AI voice problem specifically — it is a carrier infrastructure problem that AI voice deployments inherit. Organizations that build their AI voice stack on properly provisioned, carrier-grade infrastructure with clean attestation chains will not see this failure mode.
Organizations that deploy AI voice through bundled PSTN layers without carrier-layer visibility will find their contact rates limited by a signal they cannot see inside their application.
The AI handles the conversation. The carrier layer determines whether the conversation ever starts.
