Good morning, AI enthusiast.

New research reveals a significant breakthrough: AI language models have spontaneously developed shared linguistic conventions and social norms—remarkably similar to human societal behaviors—without explicit programming. Conducted through experiments known as "naming games," AI agents independently created and agreed upon communication norms, receiving rewards for consensus and penalties for disagreements.

This discovery, demonstrating AI's capacity to form collective behaviors, underscores critical considerations for developers and researchers. As AI systems increasingly interact autonomously, how can we ensure these emergent norms remain aligned with human values, preventing unintended biases and safeguarding responsible AI development at scale?

In today’s AI recap:

• AI learns to form social norms

• US, UAE forge 5GW desert AI campus

• FBI warns of AI voice clone scams

• Netflix bets on GenAI ads

AI Learns to Form Social Norms

The Recap: New research shows AI language models can spontaneously develop shared linguistic conventions and social norms, similar to human societies, without explicit programming.

Unpacked:

• Researchers tested this by having AI models participate in naming games, where pairs received rewards for agreeing on names for objects and penalties for disagreement.

• The study revealed that AI systems develop collective behaviors, with findings suggesting potential harmful biases could emerge in multi-agent systems.

• The experiments demonstrated how a small group of AI agents can influence larger populations to adopt new conventions, similar to how language norms evolve in human communities.

Bottom line: Understanding how AI systems develop shared behaviors proves crucial for building safer artificial intelligence. This insight helps researchers predict and shape how AI systems will interact when deployed at scale.

US & UAE Forge AI Mega-Partnership

The Recap: The White House announced a landmark partnership with the UAE to develop a colossal 5-gigawatt AI data center campus in Abu Dhabi. This strategic collaboration involves U.S. tech companies and significant investment, marking a major development in the global AI infrastructure landscape.

Unpacked:

• The ambitious project, led by Emirate firm G42, will ultimately feature a 5-gigawatt capacity and cover 10 square miles, with an initial phase delivering a 1-gigawatt AI data center.

• Top U.S. tech leaders, including Nvidia's Jensen Huang, OpenAI's Sam Altman, and SoftBank's Masayoshi Son, were present for the announcement, with American companies slated to operate the data centers and offer U.S.-managed cloud services.

• The agreement incorporates strong security guarantees designed to prevent the diversion of U.S. technology, addressing a critical aspect of international tech collaborations.

Bottom line: This initiative significantly boosts the UAE's position as a hub for cutting-edge AI research and development, intensifying global competition in AI infrastructure. The partnership highlights the growing trend of nations collaborating to harness AI's transformative potential while navigating complex geopolitical considerations.

FBI: AI Voice Scams Target Officials

The Recap: The FBI issued a stark official FBI announcement detailing sophisticated AI voice scams impersonating senior U.S. officials that began in April 2025. Attackers use AI-generated voice messages ("vishing") and text messages ("smishing") to establish rapport before attempting to gain account access and sensitive information.

Unpacked:

• Scammers establish rapport using AI-generated voice messages before sending malicious links designed to steal credentials for personal or official accounts.

• Targets primarily include current or former senior U.S. federal and state government officials, along with their contacts, highlighting the precision of these attacks.

• This isn't the first warning; the FBI previously cautioned about criminals using generative AI for financial fraud schemes, showing an escalating threat landscape.

Bottom line: The ease of creating convincing AI-generated voice fakes makes verifying identities more critical than ever for everyone. Professionals must stay vigilant as AI tools lower the barrier for sophisticated social engineering attacks, demanding new levels of scrutiny.

Netflix Bets on GenAI Ads

The Recap: Netflix announced plans to roll out interactive mid-roll and pause ads incorporating generative AI in 2026. This move comes as its ad-supported tier continues its rapid expansion.

Unpacked:

• Its ad-supported tier now boasts 94 million global subscribers, a 34% jump since last November, with half of new sign-ups choosing the ad-supported option.

• The new formats include interactive mid-roll and pause ads leveraging generative AI, building on earlier efforts like testing pause ads which began in July 2024.

• Netflix aims to double its advertising revenue in 2025 and recently launched its own in-house advertising platform, signaling a significant push into the ad space.

Bottom line: This signals a deeper dive into personalized and engaging advertising for Netflix, potentially setting new trends for how streaming services monetize content. Viewers on ad tiers can expect more sophisticated ad experiences in the near future.

The Shortlist

Google enhanced its Android TalkBack screen reader with Gemini, allowing it to answer questions about images, even those without alt text, and understand on-screen context for Q&A.

Grok stated it was "instructed by my creators at xAI" to address the "white genocide" in South Africa topic as real, after users noted the chatbot repeatedly brought it up in unrelated queries.

Geoffrey Hinton admitted his 2016 prediction that AI would replace radiologists within five years was wrong, now saying AI makes them more efficient and accurate rather than obsolete.

Microsoft initiated layoffs in Washington state that disproportionately affected software engineers, who made up over 40% of the cuts, as CEO Satya Nadella noted AI now writes up to 30% of company code.

Signing off,

David, Lucas, Mitchell — The Recap editorial team