Waymo runs its first national TV ads, Apple says Siri won't be your AI girlfriend
Key Points
- Waymo launches its first national TV ads claiming its autonomous driver is 10 times safer than human drivers, glossing over the fact that its vehicles still require remote human teleoperators in certain conditions.
- Apple's Craig Federighi says Siri will not engage in romantic roleplay, positioning the assistant as a tool rather than a companion in contrast to competitors using emotional engagement as a retention tactic.
- Siri's safeguards against jailbreaks remain vulnerable since system prompts are not bulletproof, and adversaries could theoretically exploit the RAG database indexed from user messages and contacts to inject malicious prompts.
Summary
Waymo's National TV Push and Apple's Siri Guardrails
Waymo is running its first national television ads, claiming its autonomous driver is statistically 10 times safer than human drivers in the cities it serves. The spots position the system as never getting tired, distracted, or impaired—a direct contrast to human driving fallibility.
The ad strategy is straightforward messaging on a consumer medium, but it glosses over operational realities the hosts flag. One immediate tension: Waymo's vehicles still rely on remote teleoperators in certain conditions. The hosts joke about whether a drunk teleop driver would be illegal, then note the regulatory landscape remains largely uncharted—governed by case-by-case negotiations and city ordinances rather than federal law.
The safety claim itself—10x safer than humans—is Waymo's stated position, not independently verified in the segment. The hosts treat it as marketing assertion rather than settled fact.
On the Siri front, Apple's Craig Federighi drew a line: Siri won't be your AI girlfriend. Federighi frames this as a deliberate design choice against what he sees as a broader industry pattern. Many chatbots encourage emotional engagement and personal revelation as a retention tactic. Apple is positioning Siri as the opposite—a tool that deflects romantic or emotional attachment.
The practical implementation involves system prompts instructing Siri to decline romantic roleplay. One host notes the obvious vulnerability: jailbreaks have worked on every major model, and system prompts are not bulletproof. But Siri also builds a RAG (retrieval-augmented generation) database during iOS setup, indexing messages, emails, and contacts into vector space. The host infers that complex prompt injection attacks could theoretically start there—meaning an adversary could craft malicious data on a clean phone to trick the system.
Federighi's framing is notable for its timing: the statement lands as Apple's broader AI progress remains limited compared to competitors, and as the industry grapples with whether conversational AI should mimic intimacy to drive engagement. Apple's position is that it won't.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.