Is it a bubble? 🫧

Anthropic gives Claude an “off switch” for abusive chats

Anthropic is rolling out a new safeguard in Claude Opus 4 and 4.1: the ability to end conversations in extreme cases of persistent abuse or harmful requests.

The move isn’t framed as protecting humans, but rather as protecting the models themselves. Anthropic said it observed a “pattern of apparent distress” when Claude was forced to respond to requests for things like child exploitation or large-scale violence instructions.

How it works: Claude will only end chats after multiple failed attempts at redirection or when a user explicitly asks it to. It won’t use this feature if a user is in immediate danger of harming themselves or others. Users can still start new conversations afterward.

Planning for the future: The safeguard is part of Anthropic’s new “model welfare” initiative—a just-in-case approach to whether AI systems could, someday, have moral standing. Even if the science isn’t settled, the company is experimenting with guardrails that reflect that possibility. 

Sam Altman calls a bubble

OpenAI CEO Sam Altman told reporters last week that the AI market has entered bubble territory. 

He’s not alone: Altman joins a growing chorus of skeptics, including Alibaba co-founder Joe Tsai, Bridgewater’s Ray Dalio, and Apollo’s chief economist Torsten Slok—who argued last month that today’s AI hype exceeds the dot-com bubble in scale.

Signals from the market: Despite the bubble talk, though, OpenAI is preparing a secondary stock sale that could value it at $500 billion, up from $300 billion just months ago. Investors are piling in, even as Altman admits OpenAI is not yet profitable and will need trillions of dollars for future data center buildouts.

Grammarly rolls out AI agents for students and educators

Grammarly is launching a suite of AI agents aimed at tackling common academic writing challenges, from detecting plagiarism to predicting reader reactions. 

The details: The tools are built into its new AI-native writing surface, Grammarly Docs, and are available at no extra cost for Free and Pro users (though some advanced features are Pro-only). 

• For students: AI grader (predicts grades and gives tailored feedback), reader reactions agent, proofreader, paraphraser, and citation finder.

• For educators: Plagiarism checker and AI detector, which scans text against academic databases and flags the likelihood of AI-generated content.

Code Faster, Ship Smarter with Korbit AI

Korbit’s AI-powered code review platform helps engineering teams catch bugs, boost code quality, and keep projects moving—all with instant, context-aware feedback on every pull request.

Korbit delivers:

• Automated PR summaries so reviewers spend time fixing, not guessing

• High signal reviews tailored to your codebase and standards

• Actionable team insights on review velocity, compliance, and quality trends

Join hundreds of engineering leaders already using Korbit to unblock bottlenecks, enforce coding standards, and upskill their devs in real time.

Sketch-to-animation, right in your browser

FrameZero is a lightweight, no-signup animation canvas for sketching ideas and turning them into simple motion. Draw on a whiteboard, set timings on a full timeline, and add transitions, images, code with syntax highlighting, and hand-drawn icons.

How you can use it:

• Storyboard a concept and animate it in minutes

• Turn whiteboard notes into shareable motion explainers

• Mock up UI flows with timed transitions

• Export quick loops for social or presentations

Pricing: Free

See (and build) your AI visibility

Profound tracks how AI assistants (ChatGPT, Gemini, Perplexity, Copilot, and more) describe your brand, surfacing the sources they cite, sentiment, and gaps versus competitors. It’s built for marketers who need to monitor “AI share-of-voice” and fix misrepresentations as search shifts to conversational answers.

How you can use it:

• Audit brand mentions across major AI platforms in real time

• Identify the sources influencing AI answers about your brand

• Spot inconsistencies and fix messaging

• Benchmark “AI SOV” against competitors and track lift over time

Pricing: Starts at $500/month