SI Agent News

• Meta's AI glasses got a privacy stress test from the German outlet heise.
• Security firm Huntress is letting IT shops trial its AI agents.
• Anthropic has filed to go public.

OpenAI halves the price of GPT-6.1 Terra

OpenAI has cut its prices in half for a new model, GPT-6.1 Terra. Terra is a mid-sized model, a middle option between small cheap models and big costly ones. The cut applies to the API (the connection that lets other apps use OpenAI's model). The Decoder reports that Terra also gains ground on OpenAI's larger models.

Decrypt says it works apps and screens more like a person would. It also reports lower prices than the earlier Terra model, which Decrypt calls GPT-5.6 Terra. Interesting Engineering, via MSN, goes further. Its headline says Terra matches a larger model, GPT-6 Astra, at a much lower price. That is one outlet's claim, so treat it as unconfirmed.

The New Stack says OpenAI launched Terra, with big gains in coding and computer-use tests. The excerpt we have does not say what the flag was. Price for everyday users: not stated. The cut is for developers.

What this means for you:

• Small business owner: ask the vendors of your AI tools whether they will pass the savings on to you.
• Job seeker or student: practice using an AI assistant on real computer tasks. That skill is getting cheaper to use.
• Creator: expect more AI features in the apps you already use, at lower cost to the app makers.

Apple ships iOS 27.2

Apple's iOS 27.2, the latest iPhone software update, is out, Digit on MSN reports. It brings a redesigned Apple Music and smarter Siri AI features (Siri is Apple's voice assistant). It also carries more than 329 security fixes across iPhones, Macs and other devices. That is nearly one fix for every day of the year, our own comparison. Price: not stated. It is worth it for anyone with an Apple device.

What this means for you: install it, then try Siri with a longer, two-part request than you normally use.

Teams to use AI against phishing

The420.in reports that Microsoft Teams will use AI to catch phishing hidden in conversations. Phishing means messages that trick you into clicking a bad link or sharing a password. Rollout details and price were not in the report.

What this means for you: keep pausing before you click links in chat. A safety net helps, but you are still the first line of defense.

Should ChatGPT see your bank account?

A Zacks piece on TradingView says to try something else before giving ChatGPT access to your bank account. It ties the idea to payment features. The excerpt does not spell out the alternative. Our version is in the technique below. Price: not stated.

What this means for you: start with a small, cleaned-up sample before you connect anything.

Meta's AI glasses get a privacy test

The German outlet heise tested how easily Meta's AI glasses (glasses with a camera and AI built in) can be used to record people in secret. It checked how well Meta's safeguards work, and says where they fail. The full test is German-language and behind a paywall, so we cannot report the details.

What this means for you: if you wear a pair, say so out loud. If someone near you does, you can ask.

Huntress opens a trial for IT firms

For strategists: Huntress, a security company, posted about a program for MSPs (firms that run IT for small businesses). They can test its agentic platform (software where AI agents take actions) and a 24/7 SOC (a team watching for threats around the clock) before rolling it out to clients. It is the company's own post, so the claims are not independently checked. Price: not stated.

What this means for you: if a small business relies on an IT firm, ask what AI watches its systems.

Anthropic files to go public

Anthropic, maker of the Claude AI assistant, filed a Form S-1 on October 26, kai-you reports. That is the paperwork for an IPO (the first sale of shares to the public), sent to the SEC, the US financial regulator. kai-you highlights the risk section, which mentions the risk of human extinction. Risk sections list everything that could go wrong, so read it as a long caution list, not a forecast.

What this means for you: filings like this are public. You can read how an AI company describes its own risks.

Broadcom helps fund Anthropic's chips

TechNews reports that Broadcom, a chip maker, is teaming up with Wall Street to raise $63 billion. The money pays for Anthropic's AI chip purchases. By our own rough estimate, that is about $190 for every person in the US. Running these models takes huge computing power.

What this means for you: that cost is why free plans have limits, and why price cuts like Terra's matter.

Terra's 'works apps like a person' skill points to one idea: the agent. An agent is a model allowed to take actions, not just answer questions.

A chatbot is like a friend you text for directions. An agent is like a driver who takes you there. It is the same brain, but now it can click, type and book.

That is why safety reports matter more as agents spread. A wrong answer is annoying. A wrong action can cost money.

Glossary term of the day: agent.

Clean sample, then ask. This is a safe way to try AI on your own spending.

1. Copy about twenty lines from a bank statement.
2. Delete your name, account number and anything that identifies you.
3. Paste the lines into ChatGPT, Copilot or Gemini.
4. Ask: 'Sort these purchases into food, travel and subscriptions. Add up each group. Flag any repeat charges.'
5. Check three totals by hand.

Price: not stated. It is worth it for anyone who dreads the monthly money review.

1. Open Settings on your iPhone.
2. Tap General.
3. Tap Software Update.
4. If iOS 27.2 appears, tap Update Now. Keep the phone plugged in.
5. When it restarts, ask Siri a two-part question.

No iPhone? Check your own phone or computer for pending updates instead.

You will know it worked when the Software Update screen says your iPhone is up to date.

When a chat message feels off, paste the text into an assistant. Remove personal details first. Ask: 'What looks suspicious here?' Try ChatGPT, Microsoft Copilot or Google Gemini.

That takes under a minute, our own estimate. Compare that with the hours it could take to recover a hacked account.

What you give up: certainty. AI can miss things, so it is a second opinion, not a verdict.

The mistake: seeing 'half price' and assuming the model suits every job.

Why it goes wrong: cheaper does not mean right for your task. Some jobs need a bigger model.

The fix: test any new model on one task you have done before and already know the answer to.

Claude (an AI assistant from Anthropic) is good at reading long documents and summarizing them in plain words. It suits something like a dense filing. Its limit: it can slip on numbers, so check key figures against the original.

Alternative: Google's Gemini.

Safe to use? Check their privacy page.

Worth it if... you often read long documents and want the gist fast.

AI models come in sizes. Small ones are fast and cheap. They suit simple jobs like rewriting an email or fixing a typo. Large ones are slower and cost more. They suit hard jobs like long reasoning or tricky analysis. Mid-sized models, like Terra, sit in between.

The rule: start small. Move up only when the answer is not good enough.

Think of a pocket knife and a full toolbox. Most days the knife is enough. Reaching for the toolbox every time wastes money and time. Learning which size fits which job is a real skill.

The term 'artificial intelligence' first appeared in a proposal for a summer workshop at Dartmouth College. Source: Dartmouth.

I asked an AI agent to clean my desktop. It did. Now I can't find my desktop.

• Explain what an agent is, and why it differs from a chatbot.
• Clean a bank sample before you share it with any AI tool.
• Match a model's size to the job instead of chasing the cheapest.

Thanks for spending a few minutes with us. If this made you smarter, share it with one person who would find it useful.