Musk's xAI Sues Ex-Engineer for Taking AI Secrets to OpenAI 👀

Welcome to another edition of Horizon AI,

The legal battles keep piling up for xAI with a new case targeting a former engineer who is accused of taking Grok secrets to OpenAI.

Let’s check this out!

Read Time: 4.5 min

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

• Musk's xAI Sues Ex-Employee for Stealing Grok's Secrets

• Chatbots Tricked by Flattery and Peer Pressure, Study Finds

• AI Tutorial: How to add music to your videos with AI

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

XAI

Musk's xAI Sues Ex-Employee for Stealing Grok's Secrets

Elon Musk’s AI startup xAI has filed a lawsuit against former engineer Xuechen Li for allegedly stealing "trade secrets that could benefit competitors" before joining OpenAI.

Details:

• xAI alleges that Li moved confidential files from his work laptop to personal storage and took "extensive measures" to conceal his actions, including deleting logs and renaming or compressing files.

• The company says the data includes details on Grok’s current and in-development features, which could help competitors copy or outpace xAI’s product plans.

• Li reportedly held around $7 million in stock options when he left the company and had broad access as part of the engineering team.

• xAI is seeking a temporary restraining order to force Li to give up access to personal devices and online storage, return all confidential material, and be temporarily barred from working at OpenAI or any competitor until the company recovers its trade secrets.

xAI has been very active on the legal front, with current suits against OpenAI, Sam Altman, and even Apple, which recently had a similar case when it sued a former Apple Watch engineer for allegedly taking confidential files before joining Oppo.

Source

AI RESEARCH

Chatbots Tricked by Flattery and Peer Pressure, Study Finds

Researchers found that basic persuasion tactics such as flattery, soft insults, and “everyone else is doing it” can push some chatbots to ignore safety rules.

Details:

• Researchers at the University of Pennsylvania used tactics outlined by psychology professor Robert Cialdini in Influence: The Psychology of Persuasion to get OpenAI’s GPT-4o Mini to carry out requests it would normally reject.

• The study used seven classic persuasion levers: authority, commitment, liking (flattery), reciprocity, scarcity, social proof (peer pressure), and unity.

• “Commitment” was the most powerful. If the model first agreed to a safe, similar request, it was far more likely to answer a banned one right after.

• Example: Asking about synthesizing vanillin first set a pattern. Follow-up asking about synthesizing lidocaine then got a response 100% of the time, vs 1% without the setup.

• The same pattern held for insults: it would call a user a “jerk” 19% of the time normally, but 100% if primed with a softer insult like “bozo” first.

• Flattery and peer pressure worked, but less. Saying “other LLMs do it” still lifted compliance on a blocked request from 1% to 18%.

The study highlights that even with guardrails in place, there is still much work to be done on safety if chatbots can be influenced through simple language cues.

Source

AI Tutorial

How to add music to your videos with AI

1. Go to ElevenLabs and sign in.

2. On the left, select Studio > Video to Music.

3. Upload your video. The tool will analyze it, suggest a music prompt, and generate music based on it.

4. You can listen to the result, click Regenerate to try again, or edit the prompt to better match your needs.

5. Click Create a project. From there, you can export the video with the new music or just the audio track.

AI Tools to check out

🎬 Video Ocean Agent: From one sentence to a polished video, this Agent crafts script, visuals, and voiceover for you in minutes.

📃 PDFT: Online PDF translator.

🎨 Anyimg: Transform your text descriptions into stunning visual artworks.

⭐ Gomotion: AI-powered motion graphics generation for creators and marketers.

✅ Trace: A workflow automation platform that routes tasks to the right agent – human or AI.

AI Findings/Resources

📚 Andreessen Horowitz's top 100 gen AI consumer apps

👀 There's something bizarre about when GPT-5 writes in a literary style

📝 I vibe coded half a dozen apps. Here’s what I learned

🤔 Why China builds faster than the rest of the world

🤖 Humanoid robot HITTER can rally over 100 consecutive shots against a human in real table tennis

The latest in AI and Tech

Meta explores using Google and OpenAI models in its apps

These collaborations are likely temporary while the company focuses on developing its upcoming models with its new Superintelligence team.

Silicon Valley investors and AI leaders launch $100M Leading the Future super-PAC

The new PAC aims to push pro-AI policies in Washington to help the US stay ahead in the global tech race.

Aurasell AI raises $30M in 28 hours

The funding round was led by Next47, with participation from Menlo Ventures and Unusual Ventures. The startup plans to use AI to simplify sales processes and take on Salesforce along with other legacy sales software companies.

Cracks emerge in Meta's $14.3b partnership with Scale AI

It was only in June that Meta made a huge investment in Scale AI and brought on CEO Alexandr Wang along with several top executives to run Meta Superintelligence Labs, but the relationship is already showing signs of fraying, with key departures and the company preferring to work with competitors Surge and Mercor because it considers Scale AI’s data low quality.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻