OpenAI Rolls Out New ChatGPT Safety Features

Welcome to another edition of Log AI.

In today’s update, we’re taking a look at two major developments: OpenAI has rolled out a new safety routing system and parental controls for ChatGPT, giving safer conversations and more oversight for teens, while Anthropic has launched Claude Sonnet 4.5, its strongest AI yet for coding and agent tasks.

Let’s get into the details.

^{Read time: 4.5 mins}

  Here's what's new today in the Log AI

• OpenAI Rolls Out New ChatGPT Safety Features

• Anthropic launches Claude Sonnet 4.5, its best AI model for coding

• AI Tutorial: How to create an AI agent with Notion in 5 minutes

• AI Tools to check out

• The Latest in AI

• ChatGPT Prompt of the day

AI NEWS

OPENAI

OpenAI Rolls Out New ChatGPT Safety Features

OpenAI has introduced a new safety routing system and parental controls for ChatGPT. These updates aim to make conversations safer and provide parents with more oversight when their teens use the app.

Key Points:

• Safety Routing: ChatGPT can now switch to a more careful model when a conversation involves sensitive or emotional topics. Instead of just refusing or giving a short reply, it tries to respond in a more thoughtful and supportive way.

• Parental Controls: Parents can link their account to a teen’s account (ages 13–17). This allows them to manage settings such as blocking image generation, setting quiet hours, or limiting voice mode. Teens also get extra protections against harmful content such as violent or sexual roleplay.

• Alerts and Notifications: If a teen shows signs of serious distress (like talking about self-harm), a trained reviewer may trigger a parent alert by text, email, or push notification. Parents won’t see the full chat history, only the alert.

• Privacy and Control: Teens can unlink, but parents will be notified if they do. Parents don’t get full access to chat transcripts, keeping some privacy in place for the teen.

• Availability: These features are rolling out now, starting with web access and coming soon to mobile apps. OpenAI also plans to use age-prediction tools to automatically apply teen-appropriate settings.

Why It Matters: The update comes as OpenAI faces pressure to improve safety for young users and after criticism that ChatGPT sometimes reinforced harmful thoughts. The goal is to give more helpful, safer responses while balancing teen privacy and parent involvement.

Source

AI AND INNOVATION

Anthropic launches Claude Sonnet 4.5, its best AI model for coding

Anthropic has just launched Claude Sonnet 4.5, its best AI model to date for coding and building agents. It’s now live and ready to use.

Details:

• Claude Sonnet 4.5 is the top performer on coding tests like SWE-bench Verified, showing stronger results than earlier models.

• It can handle long, complex tasks and stay focused for over 30 hours without losing track.

• New tools include code checkpoints (so you can roll back), a better terminal, a VS Code extension, and smarter memory for agent work.

• It’s designed to manage automation and agent tasks that mix coding, logic, and tool use.

• Pricing stays the same as Sonnet 4, starting at $3 per million tokens for input and $15 for output.

• Anthropic says this is its safest model yet, with better protection against misuse, prompt tricks, and harmful outputs.

Source

AI TUTORIAL

How to create an AI agent with Notion in 5 minutes

Creating an agent with Notion is so easy, and you don’t need coding knowledge. Here is how to do it

1. Go to Notion and make sure your workspace plan includes Notion AI.

2. Click the face/avatar icon at the bottom right, or open the Notion AI tab in the sidebar to activate the Agent.

3. Define your task or paste a prompt. For example:

   • “Compile customer feedback from Slack and Notion into a report, create a database of insights, and flag high-priority items.”

4. Create an Instructions (Memory) Page. Write down personality, working style, constraints, preferred formatting, and reference knowledge. The agent will read this before working.

5. Run the Agent Task. It will plan the steps, act within Notion (and connected tools), and deliver results such as new pages, updated databases, or reports.

6. Review the output, adjust the instructions, or refine the prompts. Re-run if needed. Over time, the memory page will help the agent better match your workflow.

Full guide with illustrations here

AI TOOLS TO CHECK OUT

• Anycrap AI: Create any absurd product with AI and pretend to buy it.

• AI Apply: Customized job applications for better results.

• Sourcetable AI: AI spreadsheet & data analyst to 10x your productivity.

• Uneversleep AI: Auto-respond to Upwork jobs with 50%+ conversion.

• Listen Hub AI: Turn any content into realistic AI podcasts.

THE LATEST IN AI

  AI recruiter Alex raises $17M to automate initial job interviews

Job seekers in all fields can expect to conduct many more initial screening interviews soon. While that may sound like positive news, it doesn’t mean that there will suddenly be more open positions.

OpenAI takes on Google, Amazon with new agentic shopping system

ChatGPT users in the U.S. can now make Etsy and Shopify purchases within conversations, marking a next step toward the future of online shopping — both for consumers and the platforms

Brave updates its AI-powered search

Brave announced on Monday that it’s adding a new feature to its AI-powered search suite, Ask Brave, to provide detailed answers on a topic based on the query.

CHATGPT PROMPT OF THE DAY

"Paraphrase the following email so that it keeps the same meaning but uses clearer, more natural wording. Keep the tone professional, concise, and polite. Do not add extra information or remove important details. Make sure the new version is easy to read and sounds human-written. Here is the email: [paste email here]."

THAT’S A WRAP!