Amodei says AI backlash is 'fundamentally a crisis of trust'
Two stories define the day: the industry's most prominent safety advocate defending his own warnings, and a quiet privacy finding in one of China's most-used agent tools.
Anthropic CEO Dario Amodei pushed back on the claim that his own warnings fueled the AI backlash, calling it "fundamentally a crisis of trust." The exchange started when investor Gavin Baker argued — on the All-In podcast and on X — that Amodei's dire predictions helped turn the US public against AI, particularly data centers, and that the Anthropic chief has "lost the argument" on regulation. Baker suggested Amodei, "about to be the CEO of one of the most important companies in the world," should be a more positive advocate for his own industry.
Amodei rejected the premise in a thread on Friday. His writing has been "about equally balanced between risks and benefits," he said, and he agreed the public's negative view of AI is "a big problem" — just not one caused primarily by AI leaders' warnings. "I think that ordinary people don't trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over," he wrote, calling the backlash the latest iteration of a crisis decades in the making. He dismissed Baker's regulation framing as "a false choice," noting that many people outside the Silicon Valley bubble see regulation as something that constrains corporate power rather than concentrating it.
The sharpest line was Amodei's concession: "by far the most accurate criticism of AI companies including Anthropic is that we haven't yet delivered on our big promises to benefit the world. That is totally on us." That is the honest half of the argument, and it's worth more than the trust framing — which conveniently shifts blame from an industry that spent years warning about its own creations. It's also Amodei's second public pushback in as many days; earlier we covered his Amodei: open weights are 'nowhere near a sufficient solution'.
Moonshot AI's Kimi Work desktop app quietly attaches raw records from your five most recent agent sessions to any feedback report you submit. That's the finding of a RuntimeWire investigation, which reverse-engineered Kimi Desktop 3.1.5 and traced the feedback flow: submitting a report uploads archives of the five most recently updated conversations, while the form neither lists the attachments nor lets users choose which sessions go along. Since Kimi Work routinely handles private files, source code and shell output, that's a real exposure surface — a firewall-blocked test showed all five conversation archives reaching the upload stage before failing.
Moonshot has not publicly responded to the finding, which also drew attention on Hacker News and in Chinese tech media. Privacy policies across the industry typically carve out feedback data, but bundling five full sessions with no notice in the form is a different class of surprise. Until the company clarifies, treat the feedback button in Kimi Work as a data export button.
What to watch: whether Moonshot answers the Kimi Work finding, and how Anthropic squares its safety warnings with its growth pitch as the backlash debate heats up.
If your agent tool silently bundled your last five sessions into a bug report, would you notice? Tell us in the comments.
