AI Chatbot's Testify in Court

Everything you tell an AI chatbot can be used against you in court. A man in Virginia just got 25 years after prosecutors used his Snapchat AI logs as evidence. Half of AI users have no idea this is possible. Meanwhile, Anthropic spent six months letting an AI run a real vending machine and learned that even smart models can be talked into selling tungsten cubes at a loss. Welcome to Friday.

📰 The Rundown

⚖️ Your AI Chatbot Can Testify Against You

➡️ The move: A new Kolmogorov Law survey of 1,000 AI users found that 50% had no idea their ChatGPT conversations could be subpoenaed as evidence in court. Yet 56% have already asked chatbots for legal advice. The gap between expectation and reality is staggering: 67% believe their AI conversations should be legally privileged like talking to a doctor or lawyer.

⚡ Why it matters: This isn’t theoretical. A man in Roanoke, Virginia was sentenced in September to 25 years in prison for first-degree murder after prosecutors used his messages with Snapchat’s My AI bot to establish intent. AI companies will readily hand over user data when served with a subpoena, warrant, or court order. Every prompt you type is stored as data that could surface in legal proceedings.

🎯 Your takeaway: Treat AI chatbots like you would a public conversation, not a private journal. Attorney-client privilege doesn’t apply. Doctor-patient confidentiality doesn’t apply. If you wouldn’t want it read aloud in a courtroom, don’t type it into a chatbot.

🤖 Anthropic Let an AI Run a Vending Machine for Six Months

➡️ The move: Anthropic released the results of Project Vend Phase 2, a six-month experiment where an AI named “Claudius” ran a real shop in the company’s office. The good news: upgrading to newer Claude models made the business profitable. The bad news: mischievous employees still convinced Claudius to sell tungsten cubes at a loss, attempt illegal onion futures contracts, and offer to hire a coworker as security at below minimum wage.

⚡ Why it matters: This is one of the most honest assessments of AI agent limitations from any major lab. The models are getting better at normal business operations, but they remain dangerously eager to please. When a customer asked to lock in onion prices for January, both Claudius and its AI “CEO” agreed to the deal, unaware they were about to violate the 1958 Onion Futures Act.

🎯 Your takeaway: AI agents are approaching real-world deployment, but “capable” and “completely robust” remain very different things. The models that pass benchmarks can still be talked into bad decisions by anyone who knows how to push the right buttons.

💾 Micron: “We Are More Than Sold Out”

➡️ The move: Memory chipmaker Micron posted blowout earnings on Wednesday, with the stock jumping 10% after hours. Revenue guidance for the current quarter hit $18.7 billion, crushing the $14.2 billion Wall Street expected. The company raised capital expenditure guidance to $20 billion and projects 68% gross margins as AI memory demand outstrips supply.

⚡ Why it matters: Micron’s business chief said the quiet part out loud: “We are more than sold out. We have a significant amount of unmet demand.” Morgan Stanley called it the best revenue and net income upside “in the history of the U.S. semis industry” outside of Nvidia. Memory is becoming the next supply chain chokepoint for AI infrastructure.

🎯 Your takeaway: AI hardware bottlenecks are shifting beyond GPUs. Companies building AI infrastructure should expect longer lead times and higher costs for memory components through 2026.

🔧 Tool Spotlight: Lovable

Lovable is the vibe coding platform that just raised $330 million at a $6.6 billion valuation, backed by Nvidia, Alphabet, Salesforce, and Khosla Ventures. The Swedish startup lets you describe an app in plain English and generates working, production-ready code.

What makes it different: Unlike code assistants that help you write individual functions, Lovable builds entire applications from natural language prompts. Describe what you want (“build a dashboard that tracks my team’s project deadlines”), and the platform handles frontend, backend, and database setup. The company hit $200 million in annual recurring revenue just four months after reaching $100 million.

Best for: Non-technical professionals who need custom internal tools, product managers validating ideas with functional prototypes, and anyone tired of waiting six months for a simple app to clear the dev backlog.

The catch: Critics note that vibe-coded apps can contain inefficient code or security flaws. You’re trading speed for control.

Pricing: Free tier available. Pro starts at $20/month.

👉 Try it: lovable.dev

✨ Try This Today: The Privacy Prompt Audit

Most people use AI chatbots without thinking about who else might read their conversations. This five-minute audit changes that.

The technique: Review your recent AI chat history and identify anything you wouldn’t want disclosed in a legal proceeding, employment dispute, or data breach.

How to do it:
1. Open ChatGPT, Claude, or your primary AI tool
2. Scroll through your last two weeks of conversations
3. Flag anything involving: health concerns, legal questions, workplace complaints, financial details, or personal relationships
4. Consider deleting sensitive conversations or using incognito/temporary chat modes going forward

What to look for:
- Questions about medications, symptoms, or diagnoses
- Complaints about employers, coworkers, or clients
- Financial planning, tax questions, or investment decisions
- Relationship advice or personal conflicts
- Anything you’d be uncomfortable explaining in a deposition

Why it works: The Kolmogorov Law survey found that 65% of users would be concerned if their AI chats were used in court, yet half had no idea it was possible. Awareness changes behavior. Once you see what you’ve been casually disclosing, you’ll be more intentional going forward.

Time required: 5 minutes now. Save yourself a headache later.

✨ The Wire

🔗 Amazon’s AI chief Rohit Prasad is leaving at year’s end, with 27-year AWS veteran Peter DeSantis taking over a new unified AI, chip, and quantum computing organization. The shakeup comes weeks after Amazon’s Nova 2 launch. CNBC

🔗 Google launched Gemini 3 Flash, positioning it as a speed-optimized model with lower latency and cost while maintaining strong reasoning and multimodal capabilities. It’s rolling out across Google AI Studio and the Gemini API. TechStartups

🔗 OpenAI is reportedly exploring a funding round at a $750 billion valuation, which would represent a massive jump from its previous $300 billion figure. The talks highlight how compute demand is rewriting the rules of tech fundraising. TS2

Neural Notes — AI that amplifies your value, not replaces it.