On October 29, 2026, CNN reported that a super PAC backing Texas Republican Ken Paxton is using a conversational AI caller to phone voters and attack his Democratic opponent, James Talarico. The PAC, Lone Star Liberty, received $10 million from Elon Musk. The caller calls itself "Jennifer," introduces itself as a "virtual staffer," and, unlike a traditional recorded robocall, is designed to ask voters whom they support and respond to what they say, according to audio recordings reviewed by CNN's KFile.
The calls are confirmed. Their reach appears small. And a pending Federal Communications Commission decision could change how widely this kind of call can be made. This post separates those three things, building on our earlier coverage of AI in the Talarico race, including the Storm-1516 deepfake video and the Google AI Overview donation mix-up.
TL;DR: what is confirmed
| Question | Answer |
|---|---|
| Who is behind the calls? | Lone Star Liberty, a pro-Paxton super PAC that received $10 million from Elon Musk (CNN, CNBC) |
| What does the AI do? | Calls voters, introduces itself as "Jennifer, a virtual staffer," asks whom they support and responds in real time |
| How was it found? | CNN's KFile identified the calls through the robocall-blocking app Nomorobo and obtained recordings |
| When? | First detected October 23, from dozens of phone numbers, reaching a relatively small number of landlines, per Nomorobo |
| What is the attack line? | One call says Talarico "took $11 million from data center special interests" |
| Is it illegal? | CNN reports no violation; the FCC's 2024 ruling covers AI-cloned voices, and the calls here use a synthetic persona |
| What could change? | An FCC petition from Club for Growth could extend political robocalls to mobile phones |
What the recordings show
According to CNN, the opening line in one recording is: "Hi, is this Julio? Hi. My name is Jennifer and I'm a virtual staffer with Lone Star Liberty calling in support of Republican Ken Paxton." The system then delivers attack lines and listens to answers.
CNN's review points to both the promise and the limits of the technology. The AI can react to a recipient, but it glitches when interrupted, stopping, restarting, and repeating questions or talking points. When asked who it supports, the caller reportedly answers: "I don't have personal opinions because I'm an AI virtual staffer. I can explain the campaign's position, but the decision is yours."
CNN also found that the same "Jennifer" voice was used by at least two other PACs this cycle. That detail matters: it suggests a commercial voice-agent product being resold to political clients, not a one-off tool built by a single campaign.
The data center claim and the irony
The line CNN highlights is that Talarico "took $11 million from data center special interests." A Lone Star Liberty spokesperson told CNN the figure was based on a Daily Caller report on donations to pro-Talarico political groups from individuals with ties to the data center industry. CNN notes those donors included Reid Hoffman, who contributed $10 million to a super PAC supporting Talarico.
That is a claim about donations to outside groups by individuals, which is not the same thing as a candidate taking money from data center companies. I have not independently checked the Daily Caller report, so read the $11 million figure as the PAC's characterization.
CNN points out the irony: the AI tool attacks a candidate over the very data center industry that makes such calls possible, while the PAC's largest donor runs a top AI company. Data centers are one of the strongest political issues of this cycle. Our coverage of the ABC and Washington Post and Ipsos poll found majority opposition, and the New York moratorium story shows how the issue is moving into law.
How the campaigns responded
Gregg Keller, a spokesperson for Lone Star Liberty, told CNN the group is "using every tool available" to inform Texas voters of their criticisms of Talarico. The Talarico campaign's statement: "Ken Paxton can't get humans to make calls for him, so Elon Musk is firing up his data centers to create the only volunteers willing to spread lies for their campaign: AI robots."
The broader fight is already personal. CNBC reports Musk gave $158 million to the Senate Leadership Fund in September, which passed the money to Texas PAC, a group backing Paxton, and separately sent $10 million to Lone Star Liberty. Talarico, who says he raised $64 million in the third quarter against $18 million for Paxton, has campaigned on the line "Texas is not for sale." The election is on November 3.
Do AI robocalls actually work?
Nobody has shown that they do. Nate Persily, a Stanford Law professor and co-director of the Stanford Law AI Initiative, told CNN he is skeptical they swing races because people rarely pick up and may hang up on a "virtual staffer." He separated the technology from the message: "manipulation is as old as politics," and the real question is whether AI is cheaper and more effective at manipulation and whether it can do so at scale.
Two facts in the reporting cut against alarm. The calls reached a relatively small number of landlines, and the system visibly glitches under interruption. Two facts cut the other way. Conversational calls can persuade one-on-one at a cost near zero per call, and, as CNN notes, this kind of outreach is hard for the public to track because there is no centralized database of campaign calls.
The FCC petition that matters more than the calls
Currently, federal law bans most robocalls to wireless phones unless the caller has prior consent, which is why these calls are mostly limited to landlines. CNN reports the FCC is considering a petition from the conservative group Club for Growth to loosen rules on political robocalls, which could permit large-scale unsolicited calls using artificial or prerecorded voices to reach mobile phones without consent. Top Democrats and consumer-rights groups oppose it, and CNN says the FCC could rule by the end of October.
This follows the FCC's 2024 declaration that robocalls using AI-cloned voices are illegal under existing federal law, after a call impersonating President Biden urged New Hampshire voters not to cast ballots in the 2024 primary. The distinction is important. A cloned voice of a real person is already covered. A synthetic persona that says it is an AI staffer is a different category, and the petition would change the channel (mobile phones) rather than the voice.
If the FCC grants the petition, a tool that today reaches landlines could reach nearly everyone with a phone. That is the number to watch, not this week's call volume.
How this fits the 2026 election-AI picture
This is the third distinct AI-and-election story in the Talarico race in under two weeks: a deepfake video attributed to Russian influence operators, a search feature that surfaced a lookalike donation page, and now an AI caller from a domestic PAC. They differ in actor and intent, which is why we keep labeling them separately. OpenAI's own midterm election war room addresses the supply side, but calls made with a commercial voice agent sit outside what any single lab can police.
What is still unknown
- How many calls were made. Nomorobo's detection covers its own users, so the real volume is unknown.
- Which vendor supplies "Jennifer." CNN found the voice across at least two other PACs but did not name a provider.
- Whether the calls comply with state law. Several states have their own AI-in-politics disclosure rules; CNN did not report on that.
- The FCC outcome. The ruling had not been made when CNN published.
- The $11 million framing. It rests on an outside report and the PAC's reading of it.
What voters can do
| Step | Why |
|---|---|
| Ask outright whether the caller is an AI | This caller answered that it was; that is a useful test |
| Do not share voting plans, personal data or payment details | Conversational callers are built to elicit answers |
| Verify claims on official sources such as the FEC and the candidate's site | Donation framings can be technically sourced and still misleading |
| Report suspicious calls to the FCC and your state election office | Complaints feed enforcement and rulemaking |
| Use call-blocking tools | Nomorobo is how this campaign was spotted |
For teams building voice agents, the lesson is design-level: disclose that the caller is AI at the start of the call, log scripts, and refuse political persuasion uses that your terms do not allow. For agent builders more broadly, AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions.
A gavel across a scroll representing the FCC petition on AI political robocalls
Bottom line
The confirmed fact is narrow and notable: a super PAC is using a conversational AI caller that discloses it is AI, against a named candidate, in a Senate race with a record-setting spend. The reach so far is small and the persuasion effect is unproven. The bigger decision is the FCC's, because it would decide whether calls like these stay mostly on landlines or move to every phone. We will add an update if the FCC rules or Lone Star Liberty or a vendor comments further.
Related reading
- Storm-1516 and the Talarico deepfake: what is confirmed
- Google AI Overview and the Talarico donation mix-up
- ABC, Washington Post and Ipsos poll on AI and data centers
- New York data center moratorium explained
- OpenAI's election war room for the midterms
Reporting as of October 29, 2026, based on CNN and CNBC; details may change as the FCC and the PAC respond.