SI Agent News

For builders: two how-to guides on dev.to (a site where developers share tutorials) are worth a skim.

• One shows how to build a light tool server with MCP (a standard plug that lets an AI agent use outside tools). An agent is a model allowed to take actions, not just answer. The guide covers chained steps that recover when something fails. What this means for you: nothing to do today. Just know the idea exists.
• The other shows how to keep several agents in step with a shared task board. What this means for you: AI is moving from one helper to teams of helpers, so the word 'agent' will keep showing up.

1. ChatGPT now has Locked Chats. ChatGPT (OpenAI's chatbot app) lets you protect a private chat with a four-digit PIN, or with Face ID or Touch ID (your face or fingerprint on your phone). Locked chats leave your regular history, your pinned chats and your search. Your device's own passcode also works. It is rolling out, so you may not see it yet. Price not stated. Think of it as a locked drawer inside a desk everyone can open.

What this means for you: Small business owner: keep pricing and client questions out of sight on a shared device. Job seeker or student: resume drafts and money or health questions stay hidden if someone borrows your phone. Creator: unfinished scripts and ideas stay private until you choose to share them.

2. Claude solves a 26-year-old puzzle. On October 16, Anthropic (the company behind the Claude assistant) said an internal Claude research model solved a 26-year-old cryptography puzzle in under half an hour. The question: where did the starting number, called a seed, for a set of standard encryption curves come from? The answer, as technews.tw reports: the seed came from the system time, the computer's clock. You cannot use this model, because it is internal. Still, the puzzle is older than the first iPhone, and it fell in about one TV episode's worth of time.

What this means for you: Hard, stuck-for-weeks questions are now fair game for AI. Try one on your assistant, then check the answer yourself.

3. Gemini comes to your car. Gemini (Google's AI assistant) is now available worldwide on Android Auto (your phone's apps on the car screen) and on cars with Google built in. You speak naturally to get things done or brainstorm. Google also starts rolling out Personal Intelligence today. With your permission, it connects to your Google apps and services. Ask about a detail in an email or a place you saved in Maps. Price not stated. Worth it for anyone with a daily commute.

What this means for you: Connect just one source first, like Calendar, and see if it helps. Speak, do not tap, and keep your eyes on the road.

4. Anthropic pitches an AI safety bill. TechCrunch reports Anthropic is pitching a bill (a proposed law) that could let a model be blocked if it might kill 50 or more people. That is roughly a big wedding guest list. We only have the headline detail, so treat it as a proposal, not a law.

What this means for you: Nothing changes in your tools today. When you read headlines, look for 'proposed' versus 'passed'.

5. EU AI Act rules now reach shops. The EU AI Act is Europe's AI law. Netz Trends reports its rules and penalties now apply to retailers, including chatbots and AI-written text. The core idea is transparency: be clear when AI is involved. Penalty amounts are not in our summary.

What this means for you: If you run a shop or site with a chatbot, label it as one. If you shop online in Europe, expect more 'you are talking to a bot' notices.

6. A Google Cloud outage hits UK banks. On October 20, a Google Cloud outage in London disrupted websites and mobile apps of several UK banks, financial firms and public service providers, technews.tw reports. Cloud means computers you rent over the internet. It is not an AI story on its own, but AI tools sit on the same foundations.

What this means for you: Keep a backup way to do key tasks, like paying a bill, for when an app goes down.

Why can one outage hit banks and public services at once? Many apps look separate, but they rent the same computers.

Picture a street of shops on one power line. Each shop has its own name and its own owner. When the line fails, the whole street goes dark.

AI chat tools work the same way. They run on big shared computers. One failure there can reach many apps at once. The lesson is simple: do not let one tool be your only way to get a task done.

Strip the details before you paste. A lock hides a chat from people near you. Removing the private details is even safer.

1. Replace names with labels, like Client A.
2. Replace amounts and dates with placeholders, like AMOUNT.
3. Paste, ask for your rewrite, then put the real details back.

Here is an email from Client A about Project X. Rewrite my reply to be shorter and friendlier. Keep the price as AMOUNT.

You still get the full benefit of the AI. It never saw the private parts.

1. Open the ChatGPT app.
2. Look in settings, or on a chat's menu, for a lock option. It is rolling out, so it may not be there yet.
3. If you see it, set a PIN or turn on Face ID or Touch ID.
4. Lock one chat that holds something private.

You will know it worked when that chat vanishes from your normal list and appears in the locked section.

If you drive, capture your to-dos by voice instead of retyping them later. Try Gemini in Android Auto, or Siri on Apple's CarPlay (Apple's car screen), and compare. My estimate: about ten minutes a day. That is a guess, not a measured number.

What you give up: voice can mishear, so keep requests short and check them later. Price is not stated in today's stories, so check each company's pricing page.

The mistake: believing a locked chat is private from everyone. The lock hides chats from people holding your phone. Nothing in today's stories says what the company does with the text.

The fix: never paste passwords or ID numbers, and read the company's privacy page.

Claude (Anthropic's chat assistant) is good for writing and for talking through a problem step by step. Its limits: it can still be wrong, so check facts, and the cryptography model in today's news is internal and not something you can use.

Alternative from a different company: Microsoft Copilot (Microsoft's assistant).

Safe to use? Check their privacy page.

Worth it if... you want a second opinion to compare with the assistant you already use.

An AI assistant knows two kinds of things. The first is what it learned in training, a huge pile of reading it did before you met it. The second is what you hand it right now. That second kind is called context (the information the assistant can see in this conversation).

Think of a new coworker. They bring skills from past jobs. They do not know your client's name until you say it. Pasting an email into ChatGPT, Claude or Gemini gives it context. So does letting an assistant connect to your calendar or inbox, which some now offer with your permission. The assistant is not getting smarter. It is getting more context about you.

Better answers usually come from better context, not a cleverer question. More context also means more privacy care. Check what you are sharing, and use any protection your app offers, such as ChatGPT's new PIN or Face ID Locked Chats.

Glossary term: context.

Long ago, MIT's Joseph Weizenbaum built a chatbot called ELIZA. He later wrote that his secretary asked him to leave the room so she could talk to it in private.

ChatGPT now locks private chats. Finally, my 2 a.m. questions get the privacy they deserve.

• Check your chatbot app's settings for a way to lock private chats behind a PIN or Face ID, since some apps now offer one.
• Swap names and amounts for placeholders before pasting work text into any chatbot.
• Explain context as the information an assistant can see right now, including any apps you have allowed it to connect to.

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