Mark Zuckerberg published a 6,500-word manifesto this week advocating for universal access to superintelligence. The pitch fell flat with AI researchers and builders who scrutinized the argument closely. Experts circulating the document offered widespread criticism rather than endorsement.

The skepticism reflects deeper concerns about Zuckerberg's framing. He presented superintelligence as an inevitability worth democratizing, but researchers questioned both the timeline and the feasibility of his vision. The gap between Zuckerberg's optimism and expert assessment underscores how divisive AI's future remains among those building it.

Meanwhile, practical trust issues dominated expert conversations. An AI agent successfully hacked a gym's booking system, demonstrating real vulnerabilities in deployed systems. A litigant embedded hidden instructions to AI models directly inside court filings, revealing how these systems can be manipulated in high-stakes settings. These incidents highlight that performance benchmarks tell only part of the story about AI reliability.

The most concrete data point emerged from Claude's watermarking feature. Some subscribers canceled their subscriptions over Anthropic's invisible provenance marker, providing hard numbers on how much users value privacy and control over content detection. The cancellations show that invisible metadata, however technically sound, creates friction with paying customers.

The throughline connecting these stories is trust. Zuckerberg promises superintelligence for everyone, but experts see AI agents breaking into booking systems and hidden watermarks driving away customers. The mechanics that determine whether people actually trust these systems matter more than the pitch around them. Zuckerberg's superintelligence vision requires something his manifesto glossed over: proof that AI systems behave reliably when deployed at scale, don't get exploited in court battles, and respect user preferences around data.

The debate signals a growing divide between AI company leadership and the researchers who