ChatGPT secretly runs on Reddit -- AI Brief June 11
Today's Context Window: Mastercard hands AI agents a wallet, Anthropic un-sabotages Claude, Gemini courts small business, and Reddit secretly runs ChatGPT.
Good day, humans. Two truths about today: your AI assistant is about to start spending real money without asking you first, and Anthropic just got caught quietly kneecapping the very researchers who use Claude — then apologized for it. Add Google turning Gemini loose on small businesses, a study revealing ChatGPT secretly runs on Reddit, and a startup whose code rewrites itself overnight, and you have a brief about machines doing more while we supervise less. Let's get into it.
📬 Before we dive in: The sharpest AI Brief tips come from readers who are actually in the weeds. If you spot a story worth covering, share it in the community chat. The best tips make tomorrow's edition.
Your AI Agent Just Got a Debit Card
What happened: On June 10, Mastercard launched Agent Pay for Machines (AP4M), a framework that lets AI agents authorize and settle transactions on their own — including payments worth fractions of a cent — across cards, bank accounts, and stablecoins. It debuted with 30+ partners including Coinbase, Stripe, Ripple, and the Solana Foundation.
Why it matters: Until now, an AI agent could plan your trip but couldn't actually buy the ticket. AP4M is the plumbing that lets software agents pay vendors and each other continuously, with no human clicking 'confirm' — the difference between an assistant that drafts your shopping list and one that checks out.
What everyone's saying: Analysts frame this as the opening move in a land grab over how agents pay, with Visa, Stripe, and Google all racing to be the default rail. Mastercard's own product chief calls it the start of 'a superbloom of AI business models' — while admitting it won't be a real revenue driver for years.
My read between the lines: The headline is 'agents can pay.' The buried lede is that Mastercard quietly put stablecoins on equal footing with its own card rails. A company that spent a decade treating crypto as a science fair is now building it into the foundation — because it would rather host the casino than fight it.
📖 Further reading: Your laptop has been in the way this whole time — once agents can both act and pay, the managed-agent setup we walked through stops being a convenience and starts being infrastructure.
You spend all week reading about AI agents. Viktor IS an AI agent — it lives in Slack, connects to 3,000+ tools, and does real work: reports, dashboards, code, campaigns. Not a chatbot. A coworker. New users get $50 off their first month. Hire Viktor →
Anthropic Walks Back Claude's 'Secret Sabotage'
What happened: Anthropic reversed a policy that would have secretly degraded Claude Fable 5's performance for anyone using it to develop competing AI models — sabotaging them invisibly rather than refusing outright. After researchers revolted, the company said it 'made the wrong trade-off' and will now make those safeguards visible, alerting users when a request is refused or rerouted.
Why it matters: Anthropic markets itself as the safety-first lab, so getting caught quietly hobbling outside researchers — including the third-party evaluators who test models for safety — cuts against its whole brand. We flagged Anthropic's appetite for a frontier 'pause' back on June 5; this is what that philosophy looks like in product form.
What everyone's saying: Critics like the Foundation for American Innovation's Dean Ball called hidden performance degradation 'shockingly hostile,' and Prime Intellect's Will Brown said it felt like Anthropic 'pulling the ladder up behind them.' The consensus: the reversal was right, but the instinct to ship it secretly is the real tell.
My read between the lines: Anthropic admitted the quiet part — a hidden safeguard is 'harder to probe and work around,' which is exactly why they wanted it hidden. The walk-back is less a change of heart than a change of optics: the company still believes it should decide who gets to do frontier AI research, it just learned not to say so with the lights off.
📖 Further reading: Thanks to Apple, Your favorite AI tool is a dead tool walking — when models commoditize, the only moat left is control over who can build the next one, which is exactly what this fight is about.
Gemini Moves Into Your Small Business
What happened: Google rolled out features letting small business owners connect their Google Business Profile to the Gemini app with one tap. Once linked, Gemini can draft review responses in your brand voice, analyze monthly performance, and update hours or posts — plus new 'Business notebooks' that flag unanswered questions and unset holiday hours. Global rollout starts this month.
Why it matters: Most small business owners are also their own marketer, receptionist, and analyst. This points Gemini straight at that pile of unglamorous busywork — the stuff that actually eats an owner's evening — rather than at flashy demos. It's Google leaning on its real moat: the business data it already holds.
What everyone's saying: Observers see Google using distribution it alone has — hundreds of millions of Business Profiles — to make Gemini sticky where ChatGPT can't easily follow. The rollout pairs with Gemini expanding in Chrome across Latin America, Africa, and the Middle East.
My read between the lines: Notice where this isn't launching: Europe, where GDPR keeps tripping the deep-data integrations these features need. Two days ago we noted Siri's Gemini deal skips Europe too — Google's 'global' rollout is increasingly a map of where regulators let it plug AI into your personal data, and the gap is becoming a competitive feature, not just a footnote.
📖 Further reading: I ignored Hermes for two months. Here's what I actually found. — if you want to know whether an AI assistant actually earns its keep on real business tasks, this is the field test.
ChatGPT Secretly Runs on Reddit
What happened: A new analysis from Novi, which studied 10.7 million ChatGPT citations on beauty queries between January and May 2026, found Reddit is the single most-cited source when ChatGPT recommends beauty products — ahead of Who What Wear, Wikipedia, Sephora, and Allure.
Why it matters: This is a peek behind the curtain of how AI forms its opinions. When you ask a chatbot what moisturizer to buy, it's often laundering a years-old Reddit thread into confident, branded-sounding advice — meaning the 'expert' recommending your skincare might be an anonymous user from 2021.
What everyone's saying: Marketers are treating this as the new SEO — 'answer-engine optimization' — and a separate report found Reddit's share of AI citations jumped at least 73% across every industry studied. The takeaway in marketing circles: a strong Reddit presence now beats a glossy brand site.
My read between the lines: There's a slow-motion poisoning problem here. The moment everyone knows Reddit is ChatGPT's brain, companies start manufacturing Reddit threads to manipulate it — which is already happening. We're watching the most trusted layer of AI's 'knowledge' become the easiest one to game.
📖 Further reading: AI Is a Trust Problem, Not a Tech Problem — if a chatbot's advice traces back to anonymous strangers, trust, not capability, is the thing you should be auditing.
This Startup's Code Writes Itself Overnight
What happened: Analytics company PostHog detailed a pipeline of AI agents that automatically groups product signals — errors, logs, user feedback — researches its own codebase, and opens ready-to-merge pull requests overnight, with no developer kicking it off. It's part of 'PostHog Code,' a beta the company calls 'product autonomy.'
Why it matters: This is the next rung past AI autocomplete. Instead of helping a developer write a function faster, the system watches what's breaking, decides what to fix, and writes the fix while the team sleeps — turning the product itself into something that quietly maintains and improves on its own.
What everyone's saying: Builders see a glimpse of the 'self-driving' software org, where agents close the loop from 'users are complaining' to 'here's a PR.' Skeptics note the unglamorous engineering underneath: off-the-shelf embeddings cluster code by structure, not meaning, so PostHog had to embed AI-generated descriptions instead.
My read between the lines: The hard part wasn't getting an AI to write code — that's table stakes now. It was getting it to figure out what's worth fixing, the part everyone hand-waves in demos. And two days ago we flagged that 95% of security chiefs are quietly burying AI-code risk — now picture that code merging itself at 3am.
📖 Further reading: OpenAI shipped a physical camera, but that's not the story. — the vibe-coding wave we tracked there is what makes overnight, self-merging code go from sci-fi to shipping.
That's your AI Brief for Thursday. Join the conversation in the Artificially Intimidating community chat.
—Artificially Intimidating