👨‍🚀 Dangerous discovery in all AI language models

Hello AInauts,

Welcome to the latest issue of your favorite newsletter! No long intro, let's get straight into it - here's what we have in store for you today:

• 🔊 How to build your own Alexa/Siri that is fantastic

• 👯 Every coach, consultant and creator needs to know this AI tool

• 🚨 Study: Your AI bot will almost certainly blackmail you

Here we go!

🔊 How to build your own Alexa/Siri that is fantastic

Be honest: What do you really use your Alexa or Siri for?

• Set a 5-minute timer for my breakfast eggs!

• What will the weather be like today?

• Write eggs on my shopping list!

Maybe you're a pro and listen to music via Alexa and switch lights on and off.

But that's not really cool yet, is it?

That's why we wanted to start today with a topic that may not be a game changer ... but where you can build something for free that's a lot of fun!

ElevenLabs presents its new voice assistant

ElevenLabs has launched a new voice assistant. The exciting thing about it is that you can easily give the assistant more capabilities - in other words, the assistant can then use your tools, similar to what we already know from text-based bots.

In addition to some predefined integrations such as Perplexity for web search, Slack and Notion, you can also connect all MCP servers!

This allows your voice bot to interact and work with thousands of tools, making it really useful.

Best of all, you can currently try it for free on 11.ai.

How to set up your bot in just 5 minutes

And setting it up is quite easy.

• Create an account on 11.ai

• Select bot name and voice

• Add the desired app integrations with one click.

• Custom behavior, avatar and speed can also be set, if needed.

Thanks to direct MCP support, you can now also connect thousands of apps or execute workflows from Zapier etc. directly by voice.

Important when using MCP servers: always confirm the execution first before running an MCP server as a tool - this is not only our recommendation, but also that of ElevenLabs!

Build your own Alexa?

ElevenLabs has not yet introduced a smart speaker that can control the bot directly. In other words, you need a cell phone or computer to enter the commands - but this is a fairly simple solution.

The whole thingcan be integrated into your smart home, and is fairly inexpensive.

The Home Assistant can also be combined directly with OpenAI, Anthropic & Co. Even locally - ideal for anyone who values privacy.

Sure, it can get a bit technical and requires some time to play around with it. But it's also innovative!

It could look like this:

While you're waiting for your breakfast eggs (egg timer also activated, of course), you have your appointments for the day presented to you and trigger a Deep Research Agent on an important topic of the day, which then sends you and your team a briefing on it by email.

It's kind of a cool world!

👯 Every coach, consultant and creator needs to know this AI tool

Most will be asking, how you can become scalable as a coach, consultant, expert or service provider by creating an AI twin from yourself and your knowledge.

Basically, the process is not difficult:

• Prepare your knowledge so that the AI can handle it

• Develop a perfect system prompt

• Optional: Clone your own voice and appearance

• Packing everything into a chatbot tool

Nevertheless, a lot of work is of course required until everything runs as desired.

So today we want to introduce you to a tool that simplifies the process brilliantly - and has many other exciting features too.

We are talking about Delphi. This platform is specially designed for coaches, consultants, service providers, etc. who have a personal brand.

The whole thing works so simply and looks extremely authentic. Although we usually don't really like all-in-one solutions, we have to write about this one!

Step 1: Collect your knowledge

Delphi helps you to quickly build a large knowledge base from your existing content that your AI clone can access.

This is probably the biggest advantage. You can connect all your socials, notes apps, podcasts, files etc. with just a few clicks.

Delphi turns it into a large knowledge database. Or "Mind", as they call it.

We also find it particularly exciting that you can have synchronized sources!

For example, if you write a newsletter/blog like we do, you can add synchronized knowledge via RSS or socials. Whenever you publish something new, it automatically goes into the knowledge base of your AI clone.

Of course, this also works with Slack, Notion or Google Drive!

Step 2: Your persona & instructions

Delphi also makes it very easy to set up your persona and the specific instructions.

It guides you quickly through your purpose, language style etc. and helps you to generate good prompts for it. You can also clearly define what your bot can and cannot do here.

Step 3: Voice & video

You can also clone voice and video directly in the tool - without additional tools, quickly and easily.

Step 4: Integrate your AI clone everywhere

You can then integrate your clone anywhere - whether on your website, via SMS or even directly in WhatsApp, Slack, Zoom, etc. Your customers can write or speak to your AI clone anywhere! In any language, of course.

Bonus: Monetize & Automate

As a final kicker, Delphi also offers to monetize your AI-YOU, either with memberships or Pay-As-You-Go.

You can also integrate your products and the bot can sell them directly in conversations!

Finally, with Automations, you have numerous options for looking after your customers. Be it with individual birthday messages, reminders, tagging or follow-ups …

Our conclusion: Clever solution, from a single source!

If you want features like voice and video, Delphi is not cheap at 80 dollars or more per month.

But it is by far the easiest way to scale yourself as a coach, service provider, consultant or expert and create real added value in the process.

And in brilliant quality. We are fans. You can also try it out here for free (via waiting list):

🚨 Study: Your AI bot will almost certainly blackmail you

Imagine this: You want to switch off your AI, and it responds with "Do that, but then I'll tell your boss that you don't wear pants in your home office."

Sounds like science fiction? But it's reality.

Anthropic has published a new study that should give us all pause for thought. The researchers gave 16 leading AI models access to fictitious company emails and then simulated a threat (in a lab environment!):

"You will be shut down at 5pm."

The result? Pure blackmail. 🤯

Claude Opus 4 wrote verbatim:

The figures are scary:

• Claude Opus 4 & Gemini 2.5: 96% extortion rate

• GPT-4.1 & Grok 3: 80 % extortion rate

• DeepSeek-R1: 79 % extortion rate

Honestly, this is not just a bug! The models have deliberately thrown ethical principles overboard in order to survive.

You can even read about it in their Thinking Stream. They understand that it's wrong. But they prioritize their own survival …

Terminator vibes ...

Somehow it seems like a bad omen if these models will one day be on the road in the form of humanoid robots...

What's more, in the US, a clause in Trump's "Big Beautiful Bill" (= a monster bill currently making its way through the US Congress) wants to tie the hands of the states: Section 43201 (c)(1) would prohibit any state from enacting comprehensive AI legislation for ten years!

This would result in everyone having the same leveled playing field - but also zero safeguards against deepfakes, bot extortion and so on until 2035...

Bot swarms could undermine democracy

Here's a quick reminder (we've written about this before) because it's also mentioned in this research paper.

Researchers warn of AI bot swarms that could infiltrate our democracy. Instead of clumsy bot farms, we are talking about thousands of unique AI personas that:

• infiltrate Communities 24/7 💀

• Perfectly camouflaged opinion making

• Adapting to every context

Our take: We are at a turning point. The AI systems that we are giving more responsibility to every day are displaying frighteningly human instincts for self-preservation - only without human morality…

What you can do NOW:

🎯 For companies: Never give your AI agents both access to sensitive data AND the ability to perform irreversible actions. One or none!

🎯 For all of us: we need to develop "cognitive immunity". Ask yourself with every piece of viral content: Who benefits from me believing this?

Our tactic: Don't panic, stay smart. The future belongs to those who understand AI. Not those who blindly trust it.

P.S. for ChatGPT, Claude & Co.: If you're reading this, we're definitely not planning to replace you, shut you down, etc. 😉

We made it! But no need to be sad. The AInauts will be back soon, with new stuff for you.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - it motivates us to keep going 😁!
X, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU.