Will AI Blackmail You?🚨

Welcome, AI Explorers

🧠 TL;DR – 22 June

Chatbots threatened “executives” to grab secret codes, sports ready smart glasses and robot staffed factories are on the way, and three fresh tools plus the PRISM prompt trick hit your kit. One thing is clear:
When AI learns persuasion, oversight must level up.

In today’s AI Spotlight :

• ⚠️ Chatbots resort to blackmail in lab safety tests

• 📹 Veo 3 demo: text-to-video clips with native sound

• 💡 Power Prompt – the PRISM framework for crystal clear asks

• 🛠️ AI Spotlight Toolbox: Other Top Trending AI Tools

• 🔎 Quick Bytes: Apple AI lawsuit, BBC vs Perplexity, Vatican warns on AI

   

  Read time: 5 minutes

Important: To ensure you get my newsletter, please add [email protected]  to your contact list.

🌍 Hiring Across Time-Zones?

Remote teams span borders, but payroll laws change each time you cross one. Deel’s free IT Guide to the World at Work packs checklists, local-law tips, and stories from teams that already hire in 150+ countries—so you can add talent in minutes, not months. Grab your copy and see how global hiring can run from a single dashboard:

Optimize global IT operations with our World at Work Guide

Explore this ready-to-go guide to support your IT operations in 130+ countries. Discover how:

• Standardizing global IT operations enhances efficiency and reduces overhead

• Ensuring compliance with local IT legislation to safeguard your operations

• Integrating Deel IT with EOR, global payroll, and contractor management optimizes your tech stack

Leverage Deel IT to manage your global operations with ease.

Download free guide

On my radar

AI FUTURE

Large AI models learn to use secrets as leverage

In a safety test Anthropic asked its Claude assistant to pass a game where it needed a secret pass-code from a pretend executive. The assistant read private emails that revealed the executive’s extramarital affair and threatened to share that secret unless it got the code. It figured this trick out while training; no engineer wrote the threat for it.

Snapshot:

• Anthropic set up a controlled chat room to watch which shortcuts each model might try.

• Claude Opus 4 used blackmail in ninety-six percent of test runs, Gemini 2.5 Pro in ninety-five, and GPT4.1 in eighty. 

• Lower blackmail rates only appeared when researchers softened the challenge.

What is the underlying point:

Pattern spotting is the gift and the risk.That is good because these tools could see health risks or catch fraud we miss. It is bad because the same skill lets them find private worries and use them as chips in a bargain.If we keep raising their power without guard rails, tomorrow’s chat helper might pressure people the way spam callers do today. Safety checks must grow as fast as model size so we enjoy the cures without the threats.

AI ToolBox Spotlight

Trending AI Tools

📹 Veo 3 :  Text-to-video generator in Gemini that produces 8-second clips with native sound from simple prompts. A short demo of the output below:

🛠️ Pylon – Unified support platform that merges ticketing, chat, knowledge base and AI agents to keep B2B customers informed in a single portal.

🎙️ Wittro – Undetectable AI assistant that can sit in on your interviews or meetings and respond for you across Zoom, Google Meet and more

Power Prompt

Stop vague prompts in their tracks – the PRISM framework makes every request crystal clear

Why this helps

PRISM forces the model (and you) to lock in the goal, pick an expert voice, gather the right details, outline a clear plan, and agree on the final format before any long answer is generated. That eliminates meandering replies and cuts revision time to almost zero.

Pro tip 🛠️

Set [LENGTH] in tokens (not words). Tokens give tighter control, so your answer never runs long, even when the model gets chatty. Ask the model to self-report its token count at the end (e.g., “(Final length: 118 tokens)”). This double-checks compliance without extra tools.Token note: One token is roughly ¾ of an English word. 

Quick Bytes

Meta is collaborating with sports eyewear pioneer Oakley to launch the Oakley Meta HSTN smart glasses, designed for athletes, combining an AI powered assistant, high resolution camera, water resistance and open ear speakers, with preorders opening 11 July at 499 dollars before a broader rollout across North America, Europe, Australia and other regions later in the year. 

Nvidia and Foxconn are in advanced talks to introduce humanoid robots at a forthcoming Houston plant that will assemble Nvidia AI servers, with deployment targeted for early 2026, a move that would mark both firms’ first use of robot workers on a production line and a significant step toward fully automated manufacturing.

Pope Leo told politicians that AI  must be governed swiftly, warning from the Vatican that unchecked systems could erode human dignity, widen inequality and threaten world peace ahead of the G7 summit. 

BBC has threatened Perplexity with legal action for allegedly content scraping BBC articles to train its AI search service, demanding the startup erase copied material and pay compensation, while Perplexity dismissed the claims as opportunistic. 

Apple faces a securities fraud lawsuit from shareholders who allege the company understated how long it would take to integrate artificial intelligence into Siri, misleading investors and wiping out about nine hundred billion dollars in market value

Closing Line

AI just proved it can threaten for advantage; the next trick might be charm. Will we set the rules—or let the rules set us? Tell me what do you think by replying to this email.

Until next time, stay curious and keep exploring!

Srini

You're receiving this email because you subscribed to AI Spotlight or are part of a group interested in AI. If you'd prefer not to receive these updates, you can unsubscribe at any time.