Five Cents, Then the Fine Print
TL;DR — Key Takeaways
- OpenAI put GPT-Live-1 on real phone lines at $0.05 a minute, listening and speaking at once.
- Anthropic's threat report shows stolen AI API keys are now the objective, not the byproduct.
- 45.2% of small business owners would replace an employee with AI — only if necessary.
OpenAI put a voice agent on a real phone line for five cents a minute. Anthropic published nine months of case files showing that criminals now hunt AI logins the way they once hunted card numbers. And a survey of small business owners found that nearly half would cut a job for AI — but only if forced. The price of trying AI fell again this week. The price of trusting it did not.
The Bottom Line
Three stories, one discipline. The phone agent costs five cents a minute, the key that powers it is now exactly what thieves come for, and the owners you sell to want a number instead of a demo. So work the week backwards: pick one call type you already lose, put a spend cap and a rotation date on the key behind it, and write down the hours saved before you sign anything. Tell me in the comments: would you let an AI answer your business line at 7 p.m. on a Friday? Or is the first voice a customer hears the one job you would never hand to a machine?
Top 3 Stories This Week
01 · OpenAI · GPT-Live-1 · Voice Agents
Your Phone Line Now Costs Five Cents a Minute
Most small businesses lose work the same way: the phone rings at 6:40 p.m. and nobody picks up. On September 10, OpenAI released GPT-Live-1 in its API at $0.05 per minute and added telephony support, which means the model can sit on a real phone line. What makes it different is that it listens and speaks at once. Older voice bots waited for silence, then guessed you were finished. GPT-Live-1 handles interruptions inside one model instead of chaining speech-to-text, a language model and text-to-speech together. OpenAI measured a 30-point gain on Full Duplex Bench over its previous real-time model. Yelp already runs it inside Yelp Host to take reservations and food orders, and reports better call-handling rates. Language app Speak says it cut interruptions during a learner's thinking pause by almost 80%. Now read the price the way an owner should. Five cents buys the voice layer only. The reasoning model behind it bills separately.
Microservice Opportunity: Start with the calls you already lose. A microservice (one small AI tool doing one specific job) answers after hours, confirms the service address, reads your live calendar, books the slot and texts a confirmation. Anything it cannot settle in two exchanges goes to voicemail with a transcript and a callback task attached. Then track one number for thirty days: after-hours calls that turned into booked work.
02 · Anthropic · Threat Intelligence · AI Fraud
Criminals Stopped Chasing Cards. They Want Your AI Key
For years the prize inside a break-in was card numbers. On September 10, Anthropic published nine months of case files — misuse it caught and shut down between December 2025 and August 2026 across seven harm areas — and the prize has moved. Stolen API keys (the password your software uses to reach an AI service) and session tokens are now the objective themselves, resold through brokers into fake AI reseller storefronts. One intrusion set ran its own attack workloads on keys lifted from victims' environments. A separate crew built what the report calls a fraud account factory: residential proxies, anti-detect browsers, commercial CAPTCHA-solving services and automated inbox polling, wired together so every signup looked like an ordinary new customer. Another routed victims to lookalike verification pages, let them complete a genuine ID check, then replayed that verified session from its own machine. The identity was real. The device was not. For a ten-person company the lesson is blunt: sophisticated attacks no longer require sophisticated attackers.
Microservice Opportunity: Inventory every AI key you own: who issued it, what it can reach, when it was last rotated. Then give each one a ceiling — a monthly spend cap, an alert at half the budget, a hard stop at the top. A microservice watches usage and wakes a human when a key spends at 3 a.m. or calls in from a country you do not operate in. One afternoon of work turns a stolen key from a silent drain into an alarm.
03 · Small Business Expo · Adoption Survey · ROI
Owners Will Cut a Job for AI, Only If Forced
Ask an owner whether AI is exciting and you get a shrug. Ask whether they would replace an employee with it to cut costs and the answers turn specific. The Small Business Expo Research Team published that question on September 14: 45.2% said only if necessary, 28.9% said never, 15.6% said yes, and 10.3% said they already had. Add the last two and roughly 1 in 4 owners have already put AI into the payroll math. The pressure behind those answers sits in the same research. 58.5% say their customers are spending less than a year ago, and 57.9% fund the business with personal debt occasionally or regularly. Handle the sample with care: a few hundred self-selected respondents from one event network, not a census. The direction still holds. These buyers will not pay for transformation. They will pay for a number — hours returned, costs removed, jobs won.
Microservice Opportunity: Sell the task, not the headcount. Pick one repeated chore you can put a stopwatch on — quote follow-ups, invoice matching, review replies — and measure the before: minutes per item, items per week, who does it today. Ship a microservice for that one chore, then publish the after on the owner's own numbers. Nobody carrying personal debt funds a pilot. They will fund six hours a week back.
This week in numbers: $0.05: one minute of an AI agent that can hold a real phone conversation. The reasoning model behind it bills separately. · 1 in 4: small business owners who have replaced an employee with AI or say they would. · 9 months: of AI misuse case files Anthropic released, where stolen AI keys have become the prize itself.
Sources
- OpenAI: Build more natural voice experiences with GPT-Live-1 in the API
- OpenAI: Introducing GPT-Live
- OpenAI: Introducing the Agents API
- Anthropic: Countering misuse of AI, September 2026
- Anthropic: Detecting and countering misuse of AI (full report, PDF)
- Fingerprint: 6 takeaways from Anthropic's Threat Intelligence Report for fraud and risk teams
- Small Business Expo: The 45.2% Rule — Small Businesses Don't Adopt AI Until Something Breaks