Welcome, humans.

Apple quietly buried a feature in the iOS 27 developer beta that would let you swap out Siri's AI brain for ChatGPT, Claude, or Gemini right from your Settings app, complete with a dedicated App Store section. The feature was never announced at WWDC. It was just...there, dormant, waiting.

Here's the awkward part: OpenAI apparently found out and is reportedly exploring legal options, including a possible breach-of-contract notice against Apple over their existing Siri partnership. So Apple is secretly building a feature that could let a competitor replace its own partner and that partner is not thrilled.

The whole thing is also blocked in the EU anyway, because Apple confirmed Siri AI won't ship there due to ongoing Digital Markets Act negotiations. So the feature nobody announced can't launch in a region that wouldn't allow it. Very normal Tuesday.

P.S. If your company reimburses professional development, reply to this email and we’ll send a one-page brief you can forward to your manager!

Here’s what happened in AI today:

…and a whole lot more that you can read about here.

Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us! 

P.S: Love robots? We’re starting a new robotics newsletter! Sign up early here.

😸 The Scam Texts Flooding Your Phone Were Built With Google's Own AI

You've seen the texts. "Your package is held." "Your toll is overdue." "Suspicious activity on your account." You probably deleted them without clicking. Smart. Because the operation behind them just got sued by Google and the numbers are staggering.

Google filed a lawsuit against Outsider Enterprise, a China-based cybercrime network accused of using Gemini (Google's own AI) to mass-produce phishing websites and flood American phones with fraudulent texts. It's the first time Google has ever sued bad actors specifically for misusing Gemini to run scams.

Here's what happened:

Why this matters: Before AI, building a convincing fake website took a coder and a few hours. Outsider Enterprise cut that to minutes, then sold the ability to anyone with $88 and a Telegram account.

The business model here is the alarming part. Outsider Enterprise also sold the toolkit to other criminals. Any non-technical buyer could log into Telegram, pay $88/week, pick a template impersonating their bank or the USPS, and start harvesting credit cards. Gemini was the production engine that generated convincing fake pages faster than any human team could.

Google is now coordinating with the FBI and has partnered with AT&T, T-Mobile, and Verizon to block the traffic. It's also backing seven bipartisan bills in Congress targeting AI-enabled scams.

Our take: The uncomfortable detail is that Outsider Enterprise used Google's own infrastructure throughout with Gemini for content, Google Cloud to host the fake sites, Google Drive to store the stolen data. Google suing them is the right call. But it's also a useful reminder that the same AI tools making your work faster are being rented out on Telegram for $88 a week to steal credit cards.

The scam texts will slow down. The toolkit still exists.

On a tangent, if you want to understand how AI models get stacked and routed under the hood, we broke down OpenRouter's new Fusion API, which lets you run prompts through multiple models at once and synthesize the best answer. Good explainer on where AI infrastructure is quietly heading. Read it here.

FROM OUR PARTNERS

The hidden cost of banking* isn’t fees. It’s navigating all your tabs and dashboards.

Every finance task takes multiple steps. Check runway? Find the right report, cross-reference your burn, interpret the graph. Pay a contractor? Track down the invoice, navigate to transfers, fill in the details. Freeze a card? Dig through settings.

Mercury Command eliminates the switching tax. It's AI built directly into your Mercury account that understands natural language and executes from one place — payments, invoices, categorization, team management. You review and approve every action. Command does the rest.

→ Try Mercury Command 

*Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., Members FDIC.

🎓 Make your agent write its own goal

So, a /goal is your AI agent’s job description for the task you want it to do: what it’s trying to do, what “done” looks like, and what boundaries it should respect. Use this in Claude Code or Codex when you want an agent to handle project-based work instead of dumping everything into a normal ChatGPT window.

Pietro Schirano (creator of MagicPath) shared a tiny agent workflow that solves a real delegation problem: before you ask the agent to do the work, ask it to define the goal itself (including for writing sub-agent goals).

His move is simple: he “basically never” writes his own /goal anymore. He asks Codex to write one for itself, plus one for each agent it spawns, meaning each helper agent gets its own target before the work begins.

The best version of this is human-reviewed autonomy: let the model draft the target, then you tighten the constraints so the agents stay on track.

As Steven Cheng pointed out in the replies, spawned agents can drift into weird edge cases without human-set boundaries, so that kinda sanity checking of the bots matters.

Try this before any multi-step agent task:

Total AI beginner? Start here (goes with this video).

Have a specific skill you want to learn? Request it here. 

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

This video is going bonkers on YouTube…

This is seriously one of the biggest surprise hits we’ve had in awhile… Jeff does a great job of explaining computers, GPUs, and the next kind of computer he’s building… there’s a reason the YT algo has been loving this one. Check it out here.

📰 Around the Horn

FROM OUR PARTNERS

Secure Your Data & AI Supply Chain: Now On-Demand

Stop patchwork tools from stalling innovation. Watch the on-demand webinar, "Elevate Data and AI Agent Security with GitLab and Google Cloud," where experts from GitLab and Google Cloud share how to secure BigQuery pipelines, automate compliance, and consolidate tooling.

Protect your code and workloads at scale—on your schedule.

Watch On-Demand

📖 Tuesday Tool Tip

So “vibe-coded” apps (made entirely with AI writing all the code) can look finished while still leaving the front door, side window, and surprisingly thick raccoon-sized crawlspace wide open.

Today’s tool tip: use AI as a pre-launch security reviewer before you put a project on the public internet. A Reddit user recently vibe-checked (hacked) a bunch of vibe-coded sites and kept seeing the same problems: no rate limits, no email verification, and API keys exposed where anyone could grab them.

So if you’re building your own software with AI, make sure you run this before launch:

Basically: treat “it works” and “it is safe to ship” as two different checkboxes.

A Cat’s Commentary

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?

P.P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.