Just in time for the beginning of Halloween week, Anthropic dropped a 318-page filing for prospective investors considering joining its much-anticipated IPO in the coming days—a bone-chilling document no matter how you slice it. If you’re an AI skeptic in the vein of Ed Zitron, who believes the bubble’s well on its way to bursting, then the company’s reported loss of nearly $81 billion in the first half of 2026 will scare your socks off. Or maybe you’re an existential-risk doomer, in which case you’ll find plenty to be afraid of in the 97-page “Risk Factors” section, featuring totally normal investor warnings such as “sufficiently advanced AI systems could pose catastrophic or existential risks to society if not developed and deployed responsibly.”
Given that I’m somebody who’s covered these companies for a while, the number one question I get from friends and family these days is about what my concrete fears look like in a world overrun by AI. Increasingly, I’ve told them that a cyberattack of some significance is at the very least plausible in the coming year. (I can’t say that I found the cybersecurity section of Anthropic’s S-1 to be terribly reassuring.) And I was saying that before the cybersecurity community declared this month “Black October” thanks to the unprecedented wave of hacking activity, including attacks on French supermarkets and Japanese financial firms, and the accidental exposure of new vulnerabilities underpinning much of the internet.
Things got even more unsettling yesterday morning, when The Washington Post published a story about how Cathedral, a new cybersecurity company founded by four ex-DOGE staffers, successfully infiltrated some of Anthropic’s internal systems in just 12 hours. The article dropped just hours before Anthropic filed its pre-IPO documentation. Never one to pass up a good Lord of the Rings reference, Palantir CTO Shyam Sankar posted, “This is an undeclared emergency. You thought you were still inside the secure bastion of the king’s castle. You have awoken in Helm’s Deep. The orc armies are coming.” Others mocked all the melodrama. “Dawg they hacked [Anthropic’s] Slack lmao chill out,” posted the popular tech-world commentator Spor.
When it comes to those of us who aren’t stuck in the B-plot of Revenge of the Nerds, how much do we really have to worry about the safety of our information? I spent my week calling up cybersecurity experts to assess the actual risk level, where AI does and doesn’t play a role, and what you can actually do today to protect yourself from the next big hack.
Signal and Noise
Tough to know where to begin with former Yahoo CEO Marissa Mayer’s deeply tone-deaf interview with The New York Times about her new “AI wife” app, Dazzle. The part where Mayer is so stumped by having to come up with three nice bullet points to describe her daughter that she has to turn to AI? The part where she suggests gender inequality in tech isn’t a major issue? I’m not sure if Mayer is stuck in the #girlboss 2010s, but I do know none of this nonsense belongs in 2026.
The New York Times convened five experts—including New York State assemblyman Alex Bores, economist Tyler Cowen, and AI Now Institute policy lead Kate Brennan—spanning the full AI-safety spectrum for a roundtable about AI regulation. The frank discussion led to some fascinating exchanges between those who fear existential risk from the technology and those who suggest the whole thing is an overblown distraction from the real issues. Read on to see which participants went from quibbling over who’s a “useful idiot” to finding common ground over the fact that labs’ internal safety pledges are “not worth the paper they’re written on.”
