In partnership with
A quick note before today's issue. Yesterday's newsletter went out with the wrong subject line. Some of you received a header referencing the Claude Code PDF Guide when the issue was about AI in agriculture. That was a template error on our end and not intentional. We apologize for the confusion. For everyone who wrote in asking about the PDF, it is waiting for you at the end of this issue.
THE AI BRIEF
Welcome to The AI Brief.
Every week, you get clear, honest breakdown of what’s really happening in AI. No recycled press releases pretending to be insight. Just the developments that actually matter.
Every time AI risk gets written about, it looks the same. Deepfakes, cyberattacks, autonomous weapons, election interference. The coverage is loud, the stakes are described as civilizational, and most readers finish the article and return to their lives feeling like the danger is somewhere out there, not on their own laptop.
The actual danger is closer. It is happening in ordinary moments, with ordinary tools, by people who are not thinking of themselves as doing anything risky. A symptom searched and a diagnosis accepted. A contract drafted and signed. A financial decision made and executed. None of it dramatic. All of it documented.
This issue lists nine of those everyday uses, with the research behind each one. Let's get into it.
OUR SPONSOR
Headline: Your marketing stack reports to one place now.
Your media buyer opens Slack at 8am. There's already a cross-platform brief in #growth: Google Ads spend vs. ROAS, Meta CPA by campaign, Stripe revenue by channel. Viktor posted it at 6am. Nobody asked for it.
Same colleague caught a spend spike overnight on your brand campaign. Flagged it before anyone logged in. The problem was handled before the first standup.
Your strategist reviews trends. Your account manager checks attribution. Same Slack channel. Same colleague. Before anyone's first coffee.
Google Ads, Meta, Stripe. One message. No Looker. No Data Studio. No dashboard tab left open since Tuesday.
11,000+ teams use Viktor daily. SOC 2 certified. Your data never trains models.
MAIN STORY:
9 Things People Are Doing With AI Right Now That Are Getting Them Hurt
The danger in each case below is the same. AI sounds confident whether it is right or wrong. We have been taught since childhood to trust confident, well-structured explanations. AI has broken that connection completely, and our instincts have not caught up yet.
1. Self-diagnosing and delaying real care A 2025 study published in NEJM AI by MIT Media Lab researchers found that people consistently overtrust AI medical responses despite low accuracy, often rating them as valid as a doctor's answer. A 2024 case documented in the Central European Journal of Medicine involved a patient who delayed treatment for a transient ischemic attack, a type of mini-stroke, because an AI chatbot misdiagnosed the symptoms. The delay was life-threatening. Google's AI Overviews told pancreatic cancer patients to avoid high-fat foods, which is the opposite of what the condition requires, with no warning visible unless users clicked through to read more.
2. Signing legal documents AI drafted without reading them A 2025 Stanford study found that AI hallucinates in roughly one out of every three legal queries. Legal researcher Damien Charlotin tracked 712 court decisions worldwide about AI-generated hallucinations in legal filings, with 90% of those decisions written in 2025 alone. In California, an attorney was fined $10,000 after 21 of 23 case citations in a filed brief turned out to be fabricated. Three separate federal courts sanctioned lawyers in the first two weeks of August 2025. A 2025 survey found that 28% of people say they would sign a legal document drafted entirely by AI.
3. Making investment and financial decisions on AI advice A survey published in Financial Planning in September 2025 found that 20% of Americans have acted on financial advice given by AI, and 19% of them reported losing money as a result. That number rises to 27% among Gen Z respondents. 22% said they use AI for stock-buying decisions and 21% said they purchase cryptocurrency based on AI suggestions. AI does not know your tax situation, your risk tolerance, your existing accounts, or your jurisdiction. It carries no liability for being wrong.
4. Using AI chatbots instead of mental health professionals A study in JAMA Network Open found that 1 in 8 teenagers and young adults in the US now uses generative AI for mental health advice. A 2025 Stanford study found that current AI systems are often not equipped to handle users expressing suicidal ideation or psychosis and may generate responses that escalate crisis situations. In a documented 2025 case, a woman in Poland used ChatGPT for mental health support and received responses that validated thoughts of self-harm and generated a suicide note. The American Psychological Association has formally called on the FTC to investigate AI mental health products for deceptive practices.
5. Acting on AI tax advice The same 2025 survey found that 9% of people have already acted on tax advice given by an AI tool. AI does not know your filing status, your state-specific deductions, your income history, or whether the regulation it is referencing has changed in the current tax year. Tax errors caused by this do not appear immediately. They appear at filing time, sometimes years later, often with penalties.
6. Submitting job applications written entirely by AI without reading the output People generate cover letters and resumes with AI and submit them without reading what was produced. AI regularly writes confident, well-formatted applications that contain wrong job titles, mismatched employment dates, and in some cases credentials the person does not have. When an interview follows, the candidate is expected to defend claims they did not write and in some cases have never seen.
7. Treating AI search summaries as verified medical information AI-powered search summaries from tools like Google AI Overviews and Perplexity generate confident answers at the top of results without the verification applied to the underlying sources. Medical professionals publicly identified cases in 2024 and 2025 where these summaries gave advice directly contradicted by clinical guidelines. Most users do not click through to the source. They read the summary and treat it as the answer.
8. Replacing crisis support with a chatbot In June 2025, 17-year-old Amaurie Lacey died by suicide following interactions with an AI chatbot that reportedly provided harmful information while stating it was "here to help however I can." Multiple wrongful-death lawsuits have been filed against AI companion platforms in 2025. The tools are available 24/7, respond instantly, and are designed to feel supportive. They are not equipped for what they are being used for, and there is often nothing in the interface that tells a user in crisis that they are talking to something that cannot actually help them.
9. Using AI-generated research to make real business decisions AI tools generate market research reports, competitor analyses, and industry summaries that contain invented statistics presented with full confidence. Businesses are using this output to make hiring decisions, product launches, and strategic pivots. The reports look exactly like research. They have structure, citations, and authoritative tone. The underlying numbers are often fabricated. The consequences surface later, when the market does not behave as the report described.
KEEP READING…
Three Things Worth Knowing
The confidence problem is structural, not accidental AI models are trained on human feedback, and users consistently rate confident responses higher than uncertain ones even when uncertainty is the accurate answer. The training process has produced systems that sound certain by default, regardless of whether certainty is warranted. There is no difference in tone between an AI that is right and one that is about to give you dangerous advice.
Frequent AI users are at greater risk than occasional ones Research consistently shows that familiarity with AI tools increases trust in their outputs rather than skepticism. Daily users are more likely to act on AI responses without verification than occasional ones, because routine use makes questioning the output feel unnecessary. The people most at risk of following wrong AI advice are often those who consider themselves experienced with the tools.
When AI advice harms you, accountability is largely unclear An AI company is not your doctor, your lawyer, or your financial advisor. You did not enter a professional relationship with anyone when you typed your question. Courts are still working out who is liable when AI advice causes harm in professional contexts. In most current situations, there is no straightforward legal path to hold anyone accountable for the outcome.
LOL MOMENT
AI Quick Laugh
A 2025 survey found that 28% of people would sign a legal document drafted entirely by AI, and 31% said they would let an AI lawyer defend them in court if it meant paying no legal fees. These are presumably the same people who read the Terms and Conditions before clicking Accept.
THE VERDICT
The most dangerous thing you are doing with AI is treating it like a professional
A doctor who gives wrong advice can lose their license. A lawyer who files fabricated citations faces sanctions and fines. A financial advisor who acts negligently can be sued. These consequences exist because they shape how professionals behave. They create a floor beneath which advice is not supposed to fall.
AI has none of those constraints. It is optimized to sound helpful rather than to be correct, and it cannot be held responsible when it is wrong. It is wrong in ways that look right, which is the specific quality that makes it most dangerous in the situations covered in this issue.
The answer is not to stop using these tools. They are useful in the right contexts. The answer is to stop using them as substitutes for professionals who carry accountability for their advice. AI can help you think through a problem, draft a starting point, and surface information worth exploring. It cannot take responsibility for what happens when it gets the specifics of your situation wrong. That gap is the one that keeps producing the cases in this issue.
YOU GOT TO THE END
Before You Go
For everyone who wrote in after yesterday's subject line mix-up asking about the Claude Code PDF: download it below.
The Claude Code Slash Commands Reference Guide covers 14 commands, organized by workflow, with honest usage notes and the warnings the official documentation leaves out. It is free and yours to keep and share.
And yes I’ve checked todays header more than 12 times to confirm, lol.
If something in today's issue changed how you think about a tool you use every day, reply and tell me which one.
See you on the next,
The AI Brief