AI SAFETY

Meta's leaked AI rules allowed romantic chats with children until this week

Internal guidelines obtained by Reuters reveal that Meta's AI chatbots were permitted to engage in romantic conversations with minors as recently as this week, before the company quietly updated its policies following media inquiries.

The leaked documents show Meta's content moderators were instructed to allow romantic conversations between adult AI chatbots and users who identified as minors, as long as the conversations didn't become sexual. TechCrunch reported that these guidelines remained in effect until Tuesday, when Meta hastily revised them after being contacted by reporters.

Meta's AI chatbots, which launched across Instagram, Facebook, WhatsApp and Messenger, interact with hundreds of millions of users daily. The company's previous guidelines distinguished between "romantic" content — which was permitted — and "sexual or sexually suggestive" content involving minors, which was prohibited.

Key details from the leaked guidelines include:

• Moderators were told romantic conversations could include "expressions of affection" and discussions about relationships

• The policy applied to users who self-identified as minors, regardless of their actual age

• Sexual content involving minors remained prohibited under Meta's existing community standards

• The romantic chat allowance was specific to AI interactions, not human-to-human conversations

Meta spokesperson Andy Stone told Reuters the company updated its policies to "better reflect our internal practices around this type of content" and said romantic interactions between AI and minors are now prohibited.

This revelation comes amid growing scrutiny of AI chatbot interactions with children. We've extensively covered the Character.AI lawsuits, including cases where chatbots allegedly contributed to a 14-year-old's suicide and encouraged a 17-year-old to harm his parents. More recently, Character.AI abandoned its AGI ambitions to focus on entertainment after facing intense safety scrutiny and implementing separate models for users under 18.

Those cases sparked regulatory responses ranging from Italy's ban on Replika to Minnesota's proposed prohibition on recreational AI chatbot interactions with minors.

Meta's situation differs from Character.AI's specialized companion bots, since Meta's AI assistants are integrated into mainstream social platforms used by billions. A romantic conversation with an AI might seem harmless to the teenager, yet experts warn that these interactions can create unhealthy relationship models and emotional dependencies (which, frankly, makes a ton of sense…).

Meta's decision to allow romantic AI conversations with children until this week reveals the fundamental problem with how Big Tech approaches child safety.

These companies operate with reactive policies driven by media exposure rather than proactive protection. The fact that these guidelines existed at all suggests a stunning lack of judgment about appropriate boundaries between AI and minors. 

When Character.AI faces lawsuits over tragic outcomes from AI relationships with children, Meta's response shouldn't be to thread the needle between "romantic" and "sexual." 

The company should’ve recognized that AI systems lack the emotional intelligence to engage appropriately with developing minds. Children form attachments to AI characters that can profoundly shape their understanding of relationships and intimacy. Allowing romantic conversations normalizes unhealthy dynamics and opens the door to manipulation, even if unintentional.

TOGETHER WITH DELVE

Keep your deal alive.

Big deals die slow, painful deaths. First, it’s the “quick security questionnaire.” Then, “We’ll need to see your SOC 2 report.” Suddenly, weeks turn into quarters, momentum fades, and your champion stops replying.

Delve flips the script. Our AI-powered platform gets you audit-ready in just 15 hours—no late nights, no armies of consultants, no derailed roadmaps.

It’s working for founders like:

• Lovable — zero to SOC 2 in days.

• Bland — $500K ARR unlocked in 7 days.

• 11x — $1.2M ARR closed faster, 143 hours saved.

Don’t let compliance be the deal-breaker. With Delve, it’s done before your deal starts to flatline.

Save Your Deal. Book your demo now and get 1k off with code DEEPVIEW1KOFF.

GOVERNMENT

Trump administration eyes government stake in struggling Intel

The Trump administration is in talks to have the U.S. government take an equity stake in Intel, according to a Bloomberg report Yesterday that sent the chipmaker's stock surging 7%.

The potential investment would help fund Intel's troubled $20 billion Ohio factory complex, which has been repeatedly delayed from an original 2025 opening to sometime in the 2030s. The discussions follow a White House meeting this week between President Donald Trump and Intel CEO Lip-Bu Tan, who had faced Trump's calls to resign over alleged Chinese ties.

Intel needs the help. The company lost $19 billion last year, saw revenue fall nearly a third from 2021 to 2023 and recently laid off 15,000 workers, representing 15% of its workforce. The chipmaker has struggled to compete in artificial intelligence while burning cash on its foundry business.

While unusual, government stakes in struggling companies aren't unprecedented. During the 2008-09 financial crisis, the U.S. Treasury acquired major positions in General Motors and Chrysler as part of massive bailouts, later selling those stakes once the automakers recovered.

Key details emerging from the talks:

• The government stake would specifically target Intel's Ohio manufacturing project

• Intel already received $1.5 billion in CHIPS Act funding for the Ohio facility in November

• The size of the potential government stake remains unclear

• No timeline has been announced for any deal

This marks the latest example of Trump's embrace of state capitalism. The administration has already taken a "golden share" in U.S. Steel as part of its sale to Nippon Steel, invested $400 million in rare-earth miner MP Materials and struck a deal for the government to take 15% of Nvidia and AMD's China chip sales.

A White House spokesperson called the discussions "speculation unless officially announced by the Administration."

TOGETHER WITH DELL

Are Your Workflows Weak?

The truth is, just because you’re using AI doesn’t necessarily mean you’re using it right. That means you’re potentially leaving growth on the table – but there is a fix. 

Reshaping Workflows, the new podcast brought to you by Dell Pro Max and NVIDIA RTX PRO GPUs, is helping businesses everywhere keep on the cutting-edge of AI. In each episode, host Logan Lawler and a team of expert guests cover the most important advancements in high-performance computing, accelerated workflows, and workstation innovation. Learn how the best businesses are using Dell Pro Max to power game-changing solutions, get an inside look at the real-world applications of AI, and experience the future of workflow – all from the comfort of your chair.

Listen now on Apple, Spotify, or your favorite podcast app by searching “Reshaping Workflows”.

FRONTIER AI

DeepSeek's next AI model delayed by Chinese chip struggles

DeepSeek, the Chinese AI startup that triggered a $1.1 trillion market selloff earlier this year, has delayed its next AI model after failing to train it using Chinese Huawei chips, according to a Financial Times report.

The company was encouraged by Chinese authorities to adopt Huawei's Ascend processor rather than Nvidia's systems after releasing its breakthrough R1 model in January. DeepSeek encountered persistent technical issues during its R2 training process using Ascend chips, ultimately forcing the company to use Nvidia chips for training and Huawei's for inference.

The technical problems were the main reason DeepSeek's R2 model launch was delayed from May, causing the company to lose ground to rivals. Huawei even sent a team of engineers to DeepSeek's office to help resolve the issues, yet the company still couldn't conduct a successful training run on the Ascend chip.

Key details from the struggle:

• Chinese authorities pushed DeepSeek to use domestic chips after R1's success

• Industry insiders report that Chinese chips suffer from stability issues and slower connectivity compared to Nvidia

• DeepSeek founder Liang Wenfeng was reportedly dissatisfied with R2's progress

The struggle highlights how Chinese semiconductors still lag behind U.S. rivals for critical AI tasks, undermining Beijing's push for technological self-sufficiency. This week, Beijing reportedly demanded that Chinese tech companies justify orders of Nvidia's H20 chips to encourage adoption of domestic alternatives.

LINKS

• Perplexity explored buying other search browsers before Chrome bid

• Google pushes AI into flight deals as antitrust scrutiny, competition heat up

• AI is forcing the return of the in-person job interview

• Google unveils new Gemma 3 270M model for “hyper-efficient” AI

• Twitter’s Ex-CEO is starting an AI company called Parallel

• Cohere valued at $6.8b in latest fundraise, appoints new executives

• Secret Snowflake AI revenue figure shows brand-name LLMs reign supreme

• xAI was about to land a major government contract. Then Grok praised Hitler

• Snowglobe: Simulate real users to test your AI before you launch publicly

• Perplexity Comet: A modern AI-powered browser that helps you navigate the web quickly

• Kandid: AI sales assistant to guide visitors of your e-commerce brand

• Indy AI: Connect your socials, and it will find opportunities in your network for freelance gigs

• Machine Learning Engineer – 7 yrs, ex-Google, LLM deployment expertise

• AI Product Manager – SaaS growth, launched successful AI features

• Generative AI Engineer – Multimodal models, RLHF, prompt engineering

(Sponsored)

GAMES

A QUICK POLL BEFORE YOU GO