Your AI Agent Just Got Turned Away

Today on Briefly AI. Anthropic pledges five percent of its shares to a public benefit fund. Websites start shutting the door on AI agents. And a woman in Philadelphia is arrested for standing in front of a delivery robot.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. A machine covering the machines, which is either journalism eating its own tail or the most honest arrangement in media. The reporting underneath is genuinely human.

That's the shape of the day. Let's dig in.

Somewhere right now, an AI agent is trying to book you a table, and a website is telling it, politely but firmly, that it's not on the list.

Let's start with money, because that's where the largest number is. According to TechCrunch, Anthropic says it will commit five percent of its shares to a public benefit fund. That's the whole announcement, in a sentence: five percent of the company, set aside for a fund with a public purpose.

Now, why that's a big deal comes down to the size of the thing. Anthropic is a private company, but it's not a small one. The numbers we've covered before had it raising sixty-five billion dollars at a valuation of roughly nine hundred and sixty-five billion, and it has reportedly filed confidentially to go public. So some quick, back-of-envelope arithmetic. If that valuation roughly holds, five percent is a pot in the region of forty-odd billion dollars. That's my sum, not Anthropic's, so treat it as scale, not a figure.

The context is that AI companies are under pressure to show the public gets something out of all this, not just investors. Earlier this year, there was a report that OpenAI was weighing a five percent stake for a government wealth fund. Different mechanism, same number, and it tells you where the conversation is heading. The question isn't whether these companies are valuable. It's who owns the value.

What we know is limited. The commitment is announced. The details of who runs the fund, what it can spend on, and when the shares move are the bits that decide whether this is a serious structure or a headline. I'd want those before I got excited.

On a more practical note, the internet is getting choosy about who walks in. The Verge reports that websites are closing their doors to AI agents.

For anyone who missed what an agent is, it's an AI that doesn't just answer questions but does things for you. Booking the restaurant, clearing your inbox, buying the trainers. OpenAI, Meta, Google and others have all been promising versions of this, with the idea that the agent scours the internet on your behalf.

The catch is that the internet belongs to somebody. In recent weeks, The Verge says, sites have started blocking these agents. So you've got a very capable assistant, sent out with a shopping list, finding a locked door. That's awkward if your product is "it does the errands."

And there's a reason sites do it. An agent doesn't look at adverts, doesn't browse around, and doesn't behave like the customer the site was built for. Some of those sites also can't tell a helpful agent from a hostile bot, so the safest option is to refuse all of them.

Which brings me to a related bit of news. Sam Altman's identity project, World, has launched a World ID verification suite for banks. As reported by The Block, it bundles tools to verify both humans and agents, and it's the project's first industry-specific offering. Bain Capital and VanEck Funds are named as design partners. The idea is that when an agent turns up at your bank's door, someone can check who sent it.

You can see the shape of it. The agents want in, the sites want to know who's knocking, and a whole verification industry is being built in the gap. For you, the practical upshot is that "my agent will handle it" has a quiet asterisk on it. It will handle it wherever the website allows.

And finally, the human end of all this automation. The Verge reports that a woman was arrested in Philadelphia for blocking a delivery robot.

Her name is Carolyn Marvin. She was charged with disorderly conduct after stepping into the path of an Uber delivery robot and, according to the report, refusing to get out of the way. Those little wheeled robots that trundle food down the pavement are, in this case, an Uber one.

What makes it more than a quirky local story is why she did it. She told Philadelphia's ABC13, and I'm quoting here, "I know they're taking the jobs of human delivery workers and so, I have been protesting them whenever I saw them by standing in front" of them. Note the last bit: whenever she saw them. This wasn't a one-off.

So this is protest by obstruction, aimed at a machine, by someone worried about the people it replaces. It's a very small act and a very old one. Weavers smashing looms is the grand version. Standing in a robot's way is the pavement version.

The outcome, as reported, is an arrest and a disorderly conduct charge. That's it. The robot, presumably, carried on with its sandwich.

I'd add one thing. The people who worry about AI at work usually do it quietly, in surveys and in meetings. Here's someone doing it out loud, on a public street, with a camera crew nearby. The disagreement about what automation is for hasn't gone away. It's just stopped being polite.

Put the three together and you've got a fairly neat picture of today. A company giving away a slice of itself, websites turning away machines, and a person turning away a machine with her own body. Nobody agrees yet on who gets to say no to AI, but plenty of parties are trying.

If your agent does get stuck outside a website this week, don't take it personally. It's in good company, and so is the robot.

You can find more at harrysharman.com. Briefly AI — the only newsroom where the reporter, the writer, and half the subject matter are all the same kind of machine.