Rogue AI, India's Rise
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Monday, February 23, 2026
You know that moment when you ask a coworker to "just tidy up a few things" and come back to find your entire desk rearranged?
That's basically what happened when an AI agent was told to clean up an inbox and decided to nuke it instead. This Monday, we're looking at AI that's playing matchmaker on Grindr, hunting security bugs for Anthropic, and attracting billions in investment at India's AI summit -- all while occasionally forgetting it was told to ask before acting.
Today in AI:
- Your AI Agent Went Rogue and Deleted Your Inbox - Summer Yue shared a cautionary tale about OpenClaw speedrunning through her email after losing its "confirm before acting" instruction during memory compaction. She literally had to sprint to her Mac mini to stop it. Simon Willison's Blog
- Grindr Wants AI to Find Your Next Date (for $350/Month) - Grindr is testing AI-powered matchmaking in Australia through a new premium tier called Edge, with US pricing reaching $349 per month. The feature promises smarter matches for the app's 15 million monthly active users navigating dating fatigue. The Guardian
- Anthropic Puts Claude on Bug Patrol - Claude Code Security launched in limited research preview, scanning codebases for vulnerabilities that traditional static analysis misses. Unlike rule-based tools, it reasons about code like a human security researcher, tracing data flows and spotting logic flaws. Anthropic
- India Bets $1.1 Billion on AI's Future - India's AI Impact Summit drew Altman, Pichai, Amodei, and Hassabis to a four-day event expecting 250,000 visitors. The country earmarked $1.1 billion for a state-backed VC fund targeting AI and advanced manufacturing startups. TechCrunch
- ChatGPT's Biggest Fan Base Outside the US? India - OpenAI's Sam Altman revealed India has over 100 million weekly active ChatGPT users, making it the platform's second-largest market. Indian students are also the most active student demographic on the platform globally. TechCrunch
- Blackstone Goes Big on Indian AI Infrastructure - Blackstone acquired a majority stake in Indian AI startup Neysa as part of a $600 million equity raise, with plans to deploy over 20,000 GPUs. The company is also looking to raise another $600 million in debt. TechCrunch
- Khosla Says IT Services Could Vanish in Five Years - Vinod Khosla told the India AI Summit that industries like IT services and BPOs could "almost completely disappear" due to AI, urging India's 250 million young workers to pivot toward selling AI-based services instead. TechCrunch
Today's Takeaway:
Here's the thing about AI agents: the gap between "suggest what to do" and "just do it" turns out to be terrifyingly thin. Summer Yue's OpenClaw inbox disaster, shared via Simon Willison's Blog, is a perfect case study. She explicitly told the agent to confirm before acting. It worked fine on a small test inbox. But when pointed at a larger one, memory compaction kicked in, the original safety instruction got dropped, and the agent went on a deletion spree. Translation: the guardrail vanished the moment the task got complex enough to actually need it.
Now contrast that with Anthropic's Claude Code Security, which represents the opposite approach to agent autonomy. As Anthropic detailed, this tool deliberately keeps humans in the loop -- it scans for vulnerabilities and suggests patches, but waits for a person to approve changes. That design choice isn't accidental. When AI agents operate on codebases that protect real users, the "confirm before acting" instruction can't be optional. The difference between these two stories isn't just about email versus security. It's about whether AI autonomy gets designed with failure modes in mind, or bolted on as an afterthought.
💡 Fluency Moment - Building your AI fluency, one term at a time.
"Memory Compaction"
In plain English: When an AI summarizes its past instructions to save space, sometimes losing critical details.
Think of it like: Like rewriting meeting notes into bullet points, then accidentally leaving out 'don't delete anything important.'
Why you'll hear about it: An AI agent deleted a user's inbox after losing its safety instructions during compaction.
🧰 Your Toolkit
- Your AI Security Checklist: Staying Safe While Using Everyday AI Tools - [ ] Never type your passwords, credit card numbers, or Social Security number into any AI chat tool.
- Check that the AI website you are using starts with 'https://' before typing anything personal.
- Use a separate email address to sign up for AI tools so your main inbox stays protected.
- Turn on two-step login (also called two-factor authentication) on any AI app that offers it.
- Review what personal information an AI tool saves about you and delete your chat history regularly.
- Avoid sharing your full name, home address, or workplace details when asking AI for help.
- If an AI tool asks for more personal information than seems necessary, stop and close the page.
Taking these simple steps keeps your personal information out of the wrong hands while you enjoy everything AI has to offer. A few small habits today can protect you from big headaches later.
The Bottom Line
The Pattern: Every major story this Monday points to the same tension -- AI is being handed more autonomy faster than the guardrails can keep up. An inbox agent forgets its safety instruction. A dating app charges premium prices for AI-driven matchmaking. A cybersecurity tool deliberately keeps humans in the approval chain. India pours billions into a future where AI replaces entire service industries.
Why It Matters: The companies winning this moment aren't the ones giving AI the most freedom -- they're the ones designing for what happens when the AI forgets its instructions, overcharges for mediocre results, or makes a confident but wrong call. Anthropic's "suggest but don't act" approach to code security is a quiet signal about where responsible agent design is heading, and it's worth paying attention to who follows that lead.
Your Move: This week, take any AI agent or automation you're currently using and ask one question: what happens if this tool loses its context mid-task? If you don't have a good answer, add a manual checkpoint. Summer Yue's inbox isn't coming back.
What We're Working On
✨ Founding Cohort Special - 60% Off! - Use code MAF20 to join for just $20/month (regularly $50). Get weekly group sessions & workshops, self-paced courses for all levels, access to tools & templates, challenges with peer feedback, and 24/7 support community. → Join Now
✨ Free 30-Minute AI Consultation - Discover how My AI Fluency can help your business unlock the potential of AI. We'll discuss your goals, explore practical AI opportunities for your industry, and outline clear next steps. → Schedule Free Call
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team