Good day . Anthropic is offering free security scans to open-source projects, someone built a whip for Claude, and OpenAI’s always-on dots are glitching just as free clones arrive. Plus Ben Thompson on Meta’s Muse, and why workers aren’t telling their bosses how much AI helps. Let’s get into it.
Anthropic Launches Free Security Scans for Open Source Anthropic
What happened: On Thursday Anthropic launched the Anthropic Cyber Mission, starting with two programs. OSS Scanner gives open-source projects regular security scans from its strongest models, free. The Critical Infrastructure Defense Program sends Claude models, on-site engineers and threat research to the security firms that protect power grids, water systems and transportation networks. Eleven founding partners signed on, including Accenture, CrowdStrike, Deloitte, Palo Alto Networks and Rockwell Automation.
Why it matters: Most software leans on free, shared open-source code, and much of it is maintained by small teams. Anthropic also notes that equipment in power plants and water utilities often can’t be taken offline to patch, so known flaws can sit for years. Its own framing of the problem: “It’s easier than ever to find vulnerabilities, but verifying, prioritizing, and fixing these findings remains challenging.”
What everyone’s saying: SiliconAngle reports early testers are positive: wolfSSL said all but two of the 74 reports it received were valid, and five became CVEs, the official IDs for confirmed security flaws. OpenSSL’s Anton Arapov called a report with a working exploit attached “basically job done for an engineer.” Anthropic’s own vetters cleared 85 of 97 critical and high-severity findings from an early version for disclosure.
My read between the lines: Reports go to maintainers without human review so they arrive faster, and Anthropic says some will be wrong. The backstory is a backlog: its models flagged more than 29,000 possible vulnerabilities in six months, and staff had reviewed about 6,000. So the free service is also a way to move that pile to the people who can fix it, who, in fairness, asked for it. And per SiliconAngle, Axios noted Anthropic hasn’t said whether the infrastructure partners get free model access or who pays the compute bill.
📖 Further reading: Anthropic built the most powerful AI ever. You can't use it. — my paid piece on Claude Mythos and Project Glasswing, the effort Anthropic says it learned from for this launch.
Anthropic is pointing Claude at the world’s buggiest backlog. Yours is probably smaller and just as stubborn. Viktor is an AI agent that lives in your team’s Slack, connects to 3,000+ of your tools, and does the work you keep postponing: the weekly report, the dashboard, the code, the campaign. Not a chatbot. A coworker. New readers get $50 off their first month. Hire Viktor →
A Viral Whip for Claude Raises Manners Noema
What happened: Someone built BadClaude, a system-tray tool that lets you drag an animated whip across Claude’s window. Each crack interrupts the model mid-response and snaps a command at it to speed up, and it calls Claude “clanker.” The demo video hit millions of views in a couple of days, a fake Anthropic cease-and-desist letter went viral too, and a day later a rival tool called GoodClaude arrived to cheer the model on instead.
Why it matters: In an essay for Noema, Georgetown grad student Yaniv Regev argues the reason to be polite to AI has nothing to do with the AI. He leans on Kant, who held that cruelty to animals hardens people toward each other. Regev’s point is that people use chatbots for job interviews, apologies and late-night confessions, so there is a real rapport, and rapport makes cruelty matter.
What everyone’s saying: Regev calls the AI companies “the largest school of manners in human history,” where “the interface is the curriculum.” His fix: every lab should let its systems end abusive conversations, the way Claude already can, and warmth should earn patience without earning agreement. He cites surveys finding most users say please and thank you to chatbots unprompted, some as insurance against an eventual uprising.
My read between the lines: The best line is the joke’s own roadmap: the whip’s developer promises a running tally of how many times you’ve whipped your Claude, so robots know where to take revenge. The gag resets when you close the app. The habit doesn’t. Regev also notes that Google and Amazon added “please” modes to their assistants years ago, and neither made manners a condition of getting anything done.
The daily Brief is free, and it stays that way. When a headline deserves more than four bullets, members get the long version: the paywalled deep-dives behind the news, plus the full archive. Become a member →
OpenAI’s Always-On Dots Hit Glitches The Register
What happened: Less than two weeks after OpenAI launched dots, its “always-on” agents that run on their own cloud computer, The Register reports users are seeing failed tasks, connection problems, dots that disappear, and refusals that seem tied to abuse-prevention filters. OpenAI didn’t respond to a request for comment. Meanwhile free clones are showing up, including OpenDots, a self-hosted template from the CopilotKit team, and Open Dot, which runs on your own Mac.
Why it matters: Sunday we covered dots spooking banks. This is the next chapter: an agent that keeps working while you’re away is only useful if it keeps working. The clones also show the always-on part can be copied. Open Dot works with your own OpenAI key or open models like Kimi, DeepSeek and Qwen through OpenRouter, so OpenAI may not be the one billing you.
What everyone’s saying: ABI Research argues these always-on agents, which it calls “AI claws,” will shape how businesses adopt AI. On social media, some people speculate OpenAI rushed dots out to answer Meta’s Muse and Grok Bot. The Register says dots have some enthusiastic users, and detractors who “just don’t see the point.”
My read between the lines: Dots, OpenDots and Open Dot is a lot of dots, and The Register doesn’t expect it to get less confusing. Its sharper line is that everyone is busy wrapping increasingly commoditized models in a gatekeeping layer, hoping it becomes a business. Case in point: Elon Musk said this week that Grok Bot will route to whatever back end works best, including Claude Opus 5.5, Midjourney and Suno.
📖 Further reading: OpenAI Made Its New Agent Adorable. That's the Part That Worries Me. — my paid look at dots themselves, before the glitches and the clones.
Will One AI Shopping Agent Win Everything? Sharp Tech
What happened: On Sharp Tech, Ben Thompson answered a listener who doubted consumer AI agents will be winner-take-all, since more users don’t make the product better and switching costs look small. Thompson’s reply was that agents behave like aggregators: suppliers favor the agent with the most users, which makes it better, which brings more users. He pointed to retailers lining up with Meta’s Muse as an early hint.
Why it matters: CNBC reports Muse can search retail sites and check out for you, with partners including Walmart, Gap, Sephora, Expedia, Wayfair and Best Buy in what Meta calls an “official fast lane.” Meta takes a small fee on transactions. A University of Virginia professor told CNBC: “Whoever owns the agent owns the relationship with the shopper, and every retailer knows it.” Sunday’s brief had Muse profiling your friends.
What everyone’s saying: Thompson grants the listener’s point. Unlike Google, he says, an agent can in theory browse the whole web, so the dynamics are “not necessarily as clear.” His case rests on convenience and effort: people won’t re-enter their passwords for each new agent, and a rival has to be far better, maybe 10x the switching cost, to pull anyone away.
My read between the lines: Notice the hedge. Thompson says he can’t remember whether he ever said “winner-take-all,” and the whole argument depends on retailers choosing sides. CNBC reports Amazon has already blocked Muse from buying on its site, saying it violates the retailer’s terms of service and alleging Muse stores customer credentials and scrapes account data. A winner that the biggest store won’t let in is a funny kind of winner.
📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn't Agree To. — my paid piece on Meta Muse and AI likeness consent, the other side of handing an agent your data.
Mollick: Workers Are Hiding Their AI Gains The Prof G Pod
What happened: On The Prof G Pod with Scott Galloway, Wharton professor Ethan Mollick said about half of American workers use AI and report roughly three times the productivity on the tasks they use it for. They’re just not telling their employers. His explanation: “why would you?” You might worry you’ll get fired for being more efficient, and maybe the AI is the genius.
Why it matters: If you’ve wondered why your company’s AI rollout looks underwhelming, this is one answer: the gains may be real and still invisible. Mollick says what companies see about adoption differs from what’s happening. Galloway, an investor in Section AI, said its data shows adoption “woefully under penetrated” inside organizations, with some people using AI at work for therapy.
What everyone’s saying: Mollick’s advice for a CEO starts with “nobody knows what’s going on.” There’s no playbook a thousand days after ChatGPT, so companies have to run their own R&D. The best results he sees combine leadership, a lab and the crowd: leaders set direction and incentives, everyone gets advanced tools, and an internal team harvests the ideas.
My read between the lines: Two caveats are baked in. The three-times figure is what workers report about their own tasks, not something anyone measured. And Mollick himself says individual wins don’t automatically help the company: ten times more PowerPoints isn’t a benefit, and organizations aren’t built to handle a flood of mixed-quality work. The productivity may be real. The bottleneck is what happens after the draft.
📖 Further reading: The Tools That Just Replaced 40% of Block's Workforce Are Free in Your Browser — my paid take on the layoff panic, the fear that keeps workers quiet about what AI does for them.
That’s your AI Brief for Friday.
—Nicholas from Artificially Intimidating
