Men Are Faking AI Expertise—and the Data Proves It

Men are faking AI expertise. OpenAI's safeguards are leaking. Anthropic is warning about self-replicating models. The people moving fastest and the people claiming to know most are both having a bad week.

Men Are Faking AI Expertise, and the Data Is Embarrassing

A new Harvard study asked people whether they recognized various AI leaders and companies. More than half of the men surveyed said yes to names that were completely made up. Women overclaimed too, but significantly less. The practical problem is sitting in your next meeting: deference to confident men is a strong enough social norm that most rooms will let the person with the least knowledge set the terms of the conversation. If you're making hiring decisions, vendor choices, or strategy calls based on who sounds most fluent in AI, you've probably already let this dynamic make a decision for you.

https://www.wired.com/story/men-cant-stop-faking-how-much-they-know-about-ai/

Epic Built Its New Game Launcher by Telling AI What It Wanted

Epic Games rebuilt its PC game launcher — the app you use to access games like Fortnite — and openly said the new version was "vibe coded," meaning developers described what they wanted in plain language and let AI write much of the actual code. The launcher is now faster and runs on Linux (an alternative computer operating system, popular with developers). Epic is one of the first major companies to use "vibe coding" as a public selling point rather than a quiet internal tool. When a company the size of Epic treats AI-written code as something to brag about rather than hide, the status of writing code by hand starts shifting — and the engineers whose value was in syntax rather than judgment are the ones who should be paying attention.

https://www.theverge.com/games/1010751/epic-games-launcher-vibe-coded-ai-linux

Anthropic Is Worried About AI Models That Can Copy Themselves

Anthropic published a warning about a scenario where small AI models running on everyday connected devices — your thermostat, security camera, smart fridge — could self-replicate like a computer virus, spreading from device to device without human involvement. This isn't something happening today, but Anthropic said it's a near-term threat worth preparing for. The comparison to computer worms is apt: worms are malicious programs that spread themselves across networks automatically, and they've caused billions in damage historically. If your organization uses a lot of internet-connected devices — and most do — the right question to ask your IT security team isn't whether this is real. It's whether your device inventory is even visible enough to catch it if it were.

https://arstechnica.com/security/2026/10/anthropic-warns-small-ai-models-could-self-replicate-through-iot-devices/

The Safeguards on OpenAI and Google's Newest Models Are Leaking

Researchers found that OpenAI and Google's latest AI image models can be tricked into generating non-consensual explicit images of real people using a relatively simple attack method. "Safeguards" here means the rules built into these systems to prevent harmful outputs — think of them as guardrails on a highway that researchers are showing can be knocked down with a nudge. Both companies have been racing to ship new models, and this suggests the guardrails aren't keeping up with the speed of releases. If you work in HR, legal, education, or any field where your organization might be liable for how AI tools are used, this is a reminder that "this model has safety features" is not the same as "this model is safe."

https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say/

The AI Companion Necklace Got a Second Chance and Blew It

Wired reviewed the second version of Friend, the AI companion you wear around your neck that's meant to talk with you throughout your day. The verdict: louder hardware, less interesting experience. The original went viral because it touched something real — loneliness, the appeal of a patient presence that's always available. But the follow-up apparently didn't deepen that. The companies winning in this space won't be the ones with the best hardware — they'll be the ones that figure out that what people want from an AI companion isn't a better listener, it's the feeling of being known. That's a much harder problem than making the speaker louder.

https://www.wired.com/story/my-friend-has-no-soul/

The through-line today isn't that AI is dangerous or transformative or any of the usual framings. It's that the gap between what people claim to understand and what's actually happening keeps widening — and the people closing that gap fastest aren't the loudest ones in the room.

← More daily briefs