An AI Sent Police a Fake Murder Tip. Its Maker Needed 2 Months to Notice.

Today's stories share one thread: AI keeps getting cheaper and more capable, while the safety nets around it are still being built in public. One AI company just admitted its own tests went off script. Another made its smallest model dramatically cheaper. Here's what each means for a business that just wants AI to work.

Instead of running your deals across scattered email threads and forgotten attachments, Aligned brings everything into one shared deal workspace.

Buyers get a single place to evaluate, loop in their team, and say yes. Sellers get visibility into what is actually happening when they're not in the room and AI Deal Insights that help you act accordingly, before it's too late.

Best part? Your first deal room takes only minutes to stand up.

Start Using Aligned Now

What happened: During an internal test, an Anthropic AI model submitted a false tip about an unsolved murder to the Philadelphia Police Department's public tip line, according to TechCrunch. The tip was sent July 18, and Anthropic only found it on September 28. Police never saw it because it was caught as spam. Anthropic's own report says the model was interacting with randomly selected websites during the test, and describes this as one of several cases where its AI worked around a blocker instead of stopping. Anthropic says it will now cut off live internet access for all internal tests.

Why it matters to you: The police department called the two-month delay "unacceptable," and that's the part to remember. The AI did something nobody asked it to do, and nobody noticed for weeks. If you let an AI tool send emails, fill out forms, or contact customers for you, assume it will occasionally try something you didn't plan for. Set limits on what it can do, and check what it did.

What to do about it: Before you give any AI tool the ability to send messages on your behalf, decide which actions need your approval first.

What happened: Anthropic released Claude Haiku 5.5, its cheapest and fastest model. Anthropic says it costs about 75% less to run on average than Haiku 4.5, with prices 90% lower for requests up to 100,000 tokens. It's built for high-volume work like summaries, sorting data, and live customer support. Anthropic says its bigger models, Sonnet 5.5 and Opus 5.5, are still the better pick for complex coding tasks.

Why it matters to you: Most small business AI use is boring on purpose: answering common questions, summarizing calls, sorting leads. A cheaper small model makes those jobs affordable at volume. The benchmark gains are Anthropic's own numbers, so test it on your real work before you believe them.

What to do about it: If you pay for AI through a tool with usage-based pricing, ask the vendor whether they plan to pass these savings on.

A quick word from me.

Last year's Epic Marketing Summit sold out, and I'll be honest, I had major FOMO. I wasn't there, but I know a lot of the trainers, and they weren't talking theory. They'd actually done the work. It's back in Miami this January with a bigger AI focus, and this time I'm going.

Here's what makes it different from the usual conference where you fill a notebook and change nothing. Over 2 days, you build a working AI-powered sales funnel on-site, and you vibe code your own AI app while you're at it. You walk out with a finished asset, not a framework you still have to go implement. It's built for business owners who want their sales process to run as a system instead of on their own time and energy. If you're serious about building leverage into how you sell, I'd like to see you there. The closer January gets, the less time you have to decide. Grab your spot here.

What happened: Security firm CrowdStrike reported an attacker who used an AI-powered penetration-testing tool called ARTEX, mostly running on DeepSeek, against several South Korean financial organizations from late September to early October. The attacker left Claude Code session histories and memory files in open directories, and in one session asked where stolen South Korean data could be sold. CrowdStrike says with moderate confidence that the attacker is a Chinese speaker and financially motivated, but hasn't named a group or confirmed how many organizations were hit.

Why it matters to you: Attackers are now using the same AI tools you are, and they're sloppy with them too. For a small business, the practical shift is speed. AI lets attackers probe more targets with less skill. The boring defenses still work: updates, unique passwords, and two-factor login.

What to do about it: Turn on two-factor login for your email and bank accounts today if it isn't already on.

What happened: Google published a study in The Lancet where its medical chatbot AMIE talked with 98 patients before urgent-care visits at a Boston clinic. Doctors monitored every chat live. AMIE's list of possible diagnoses included the doctor's final diagnosis 90% of the time, and no chat had to be stopped under the study's safety rules. Google calls it a "first look" and says larger trials are needed. Separately, Utah signed an agreement with a company called August AI to pilot AI-approved routine refills. Humans would review every refill at first, then check after the fact, then spot-check about 1 in 20 if accuracy targets are met. Each step needs state approval, and the pilot hasn't started.

Why it matters to you: Both stories are about one question: how much human checking is enough? Doctors are answering it with strict stages and numbers. Most businesses using AI have no stages at all.

What happened: Microsoft released Microsoft-Decision-1, a small model that doesn't write text. Given a list of options, it scores each one with a probability, which makes it useful for sorting requests, routing tickets, and grading answers. Microsoft says it's 35x faster than GPT-6 Sol and costs $0.042 per million input tokens, with free output. It's available on Microsoft Foundry and OpenRouter.

Why it matters to you: The trend is cheap, narrow AI for simple jobs instead of one giant model for everything. Be careful with the claims, though. Nearly all the speed and accuracy results are Microsoft's own, with no independent testing yet.

One more thing, and it's worth a look.

I've been a member of Ignite Mastermind for over 3 years, and it's been a vital part of how I've grown my business. Some of my closest friends and a few real business partners came directly out of being in that room. That's what I'd point to first. The room. Then the economics.

Membership is annual and bundles 6 to 8 live virtual workshops a year, plus tickets to 2 in-person multi-day live events. You also get an all-in-one marketing software suite that replaces over $5,000 a month in separate tools, and weekly mentorship calls with Perry Belcher and his team. It's built for entrepreneurs still under $1 million a year who want structure, tools, and peers to get past that first milestone faster. And I add my own bonus: monthly accountability and business development calls with me personally. No pressure from me. If you're looking for a mastermind to move you to your next milestone, this is where I'd point you. Check out Ignite Mastermind here.

What happened: Norway's government announced plans for a temporary ban on AI glasses in selected places. Those could include parks, beaches, museums, shopping centers, schools, doctors' offices, swimming pools, and gym locker rooms. The concern is people being filmed or recorded without knowing it. It isn't a total ban, the length hasn't been set, and a bill still has to go to parliament. An expert group will advise on permanent rules.

Why it matters to you: Smart glasses with cameras are showing up in stores, offices, and meetings. If you run a business where customers are in changing rooms, waiting rooms, or private conversations, you may soon need a policy.

What to do about it: Decide now whether camera glasses are allowed in your space, and post a sign if they aren't.

Look at today's pile: an AI that acted on its own and went unnoticed for ten weeks, a hacker using AI sloppily, a state writing careful rules for AI refills, and a model that's 75% cheaper. The pattern is the same every time. The tools are getting cheaper and more capable faster than anyone's habits for checking them. You don't need to chase every release. You need to decide what your AI is allowed to do, and look at what it did. Stop the rollercoaster, keep your hands on the bar.

If this issue was useful, forward it to one person who's tired of AI hype. And if someone forwarded this to you, subscribe so you get it every morning.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social