Trust Signals You Rely on Can Be Faked in Seconds

Hey Everyone - I watched a video this week that made my stomach turn. A telehealth company called Medvi reached a $1.8 billion valuation. The New York Times wrote a glowing profile. And then someone dug into the details and found fake testimonials built from stock photos, AI-generated doctor profiles using real people's names without consent, and products that a class-action lawsuit alleges do not actually work. The company even received an FDA warning letter.

This one hit differently for me because AI made every part of it scalable in a way that was not possible a few years ago. That has implications for all of us.

This week:

Let's dive in.

A telehealth company called Medvi recently hit a $1.8 billion valuation. The New York Times covered it as a success story of AI-driven efficiency in healthcare. Then a creator (Coffeezilla) dug into the company and what he found was ugly.

Fake before-and-after weight loss testimonials built from stock photos and images pulled from unrelated people on the internet. AI-generated doctor profiles in their advertisements using the names of real medical professionals without their knowledge or consent. A core product, oral tirzepatide, that is the subject of a class-action lawsuit alleging it has no proven clinical efficacy. An FDA warning letter for false and misleading claims. Customer reviews describing charges for products that were never delivered and an inability to get refunds.

This is not a story about one bad company. It is a story about what becomes possible when AI makes deception cheap and scalable.

Every one of those fraudulent practices existed before AI. People have been faking testimonials and making misleading health claims for as long as commerce has existed. But each of those things used to require effort. You had to find real-looking photos. You had to write convincing copy. You had to build a website that looked legitimate. You had to do it all at a scale that was limited by how many hours were in the day.

AI removed those constraints. You can generate realistic testimonials in seconds. You can create fake professional profiles that look completely authentic. You can build a polished, trustworthy-looking website in an afternoon. You can produce marketing content at a volume and quality that used to require an entire team. The cost of looking legitimate has dropped to nearly zero.

That is a problem for everyone, not just the people who get scammed by companies like Medvi.

If you are a founder building something real, you are now competing against people who can fabricate credibility at scale. Your actual testimonials from actual customers look identical to someone else's AI-generated ones. Your real credentials sit next to fake ones that are indistinguishable on a screen. The honest version of your business and the fraudulent version of a competitor's business look the same to a consumer scrolling through search results.

If you are a consumer, your ability to tell the difference between real and fake is getting worse, not better. The signals we used to rely on to evaluate trustworthiness are being replicated perfectly by people who have no intention of delivering on their promises. A professional website used to mean something. A doctor's name on a product used to mean something. A before-and-after photo used to mean something. All of those signals can now be manufactured.

I do not think the answer is to stop trusting everything online. That is paralyzing and impractical. But I do think we need to get much more deliberate about where we place our trust and why.

The businesses and people who will thrive through this are the ones who invest in forms of trust that are hard to fabricate. In-person relationships. Verified track records. Community reputation built over time through consistent delivery. Word of mouth from people you actually know. These are the signals that AI cannot easily replicate because they require real human interaction and real time and real follow-through.

The Medvi story is a preview of what is coming across every industry. As AI makes the surface-level indicators of trustworthiness easier to fake, the deeper indicators become more valuable. If you are building something legitimate, that is actually good news for you in the long run. But only if you are investing in the kind of trust that cannot be generated by a prompt.

(Here is the video that started this rabbit hole for me: https://www.youtube.com/watch?v=0A2SP-QBByI)

I have been thinking a lot lately about what I am actually building with Austin Founders Feed, and I want to be precise about it. I am not building a newsletter. I am building a media asset.

The distinction matters. A newsletter can be a hobby, a side project, a personal brand extension. A media asset is a business that has value independent of the person who created it.

Think about some of the exits in the newsletter space. The Hustle sold to HubSpot. Morning Brew sold to Business Insider. Industry Dive sold for $525 million. Axios sold for $525 million. These were not personality-driven brands that lived and died with one creator. They were media companies with audiences, revenue streams, editorial systems, and brand equity that could transfer to a new owner.

That is fundamentally different from what most creators are building. If you are building a YouTube channel or a podcast, you are in most cases building around your face and your voice. That is not a bad thing. It can be incredibly lucrative. But it is hard to sell a channel that is inseparable from the person on camera. The audience came for you. If you leave, the audience leaves.

Newsletters do not have this constraint. The audience subscribes for the content, the curation, the perspective on a specific topic. The format does not require my face. It does not require my voice. It requires a clear editorial point of view, consistent quality, and a real connection to the audience. All of those things can be documented, systematized, and eventually handed off.

This is how I am thinking about every decision I make with Austin Founders Feed right now. The business directory, the sponsors page, the guides, the events. Each of these is a feature of the media asset, not just something fun to add. Each one increases the value of the business independently of me.

When I build a newsletter, I am building a media company. When I do YouTube, I am building a channel. Both can make money. But only one of them is an asset I can eventually sell, scale beyond myself, or hand to someone else to run.

That framing changes everything about how I prioritize my time.

Last week, Anthropic accidentally shipped the entire source code of Claude Code in a public npm package. A debug source map file that should have been excluded from the release pointed to a zip archive on Anthropic's cloud storage. Publicly accessible. Within hours, all 512,000 lines of TypeScript were mirrored on GitHub and forked over 41,500 times before Anthropic could take it down.

Anthropic called it "a release packaging issue caused by human error, not a security breach." The cause was almost embarrassingly simple. Someone forgot to add a single line to a config file that would have excluded the source map from the published package.

The security drama is interesting but what I found much more instructive is what the code actually revealed about how the most valuable AI product in the world is built. Claude Code is reportedly generating $2.5 billion in annualized revenue. And the source code shows that the competitive advantage is not some secret model trick.

It is the system around the model.

Claude Code does not operate like a single AI assistant. Under the hood, it breaks complex problems into smaller tasks and delegates them to worker agents. Those agents are managed through system prompt instructions like "do not rubber-stamp weak work" and "you must understand findings before directing follow-up work." The memory and context management is sophisticated. There are 44 hidden feature flags for capabilities that are fully built but not yet shipped. There is an unreleased autonomous daemon mode called KAIROS that can persist across sessions and independently decide when to act.

The takeaway for builders is clear. The future of AI is not just better models. It is better systems around those models. The people and companies winning with AI right now are not the ones with access to some secret intelligence. They are the ones who have thought carefully about memory, context, workflow orchestration, and guardrails. The model is the engine. The system is the car.

And one more thing. This entire leak happened because of one missing line in a config file. A company valued at tens of billions, generating billions in revenue, with some of the best engineers on the planet, exposed their most valuable intellectual property because of a single human oversight. That is a good reminder that no matter how sophisticated the technology gets, the human details still matter enormously.

Here are two good deep dives if you want to learn more:

If the Signal section this week made you uneasy, good. It should. AI is making it trivially easy to look legitimate online. Which means looking legitimate is no longer a reliable signal of actually being legitimate.

So the skill is building trust through channels that are hard to fabricate.

The first is physical presence. Show up in person. Attend events. Shake hands. Have real conversations with real people in real rooms. None of this can be faked by an AI. When someone has met you, looked you in the eye, and had a conversation with you, that creates a form of trust that no website or testimonial page can replicate. This is one of the reasons I keep coming back to the value of local community. The more the digital world fills up with noise and deception, the more the physical world becomes a trust advantage.

The second is a verifiable track record. Do things in public. Ship work that people can see. Document your results honestly, including the failures. Over time, a body of visible, consistent work becomes its own form of proof. It is very hard to fake years of publicly documented effort. Someone can fabricate a single impressive case study. They cannot fabricate a three-year history of weekly newsletters, community events, real partnerships, and public conversations with real people who will vouch for them.

The third is word of mouth from real relationships. When someone you trust tells you "I worked with this person and they delivered," that carries more weight than a hundred five-star reviews from strangers. Building a business where your customers actively refer you to others is the most durable competitive advantage in a world where online social proof is becoming unreliable. This takes longer to build. It is harder to scale. That is exactly why it works.

The fourth is consistency over time. Trust compounds. Showing up once is easy to fake. Showing up every week for two years is not. The businesses and people who will stand out as AI-generated content floods every channel are the ones who have been present, consistent, and reliable long enough that their reputation speaks for itself.

None of these are new ideas. People have been building trust this way for centuries. What is new is that the alternative paths to appearing trustworthy are being automated and cheapened by AI. The shortcuts are getting shorter. Which makes the long road more valuable than ever.

Invest in the forms of trust that require real time, real effort, and real human connection. Those are the only ones that will hold up.

Three questions I am sitting with this week:

🤖 Weekly AI Prompt

"I run [describe your business]. Help me audit my online presence for trust signals. What would a skeptical potential customer look for to verify that my business is legitimate? Identify any gaps where my credibility depends entirely on things that could be easily faked, and suggest ways I can add verifiable, hard-to-fabricate trust signals."