📰

subtl daily briefing

Share𝕏in

Good morning, founders and builders. The AI industry just got a reality check as every frontier model failed a new benchmark designed to test genuine intelligence, while Meta and YouTube lost landmark social media addiction trials that could reshape Big Tech liability. Meanwhile, geopolitical tensions continue disrupting global markets as the Iran conflict pushes oil prices toward stagflation levels.

In today's briefing

  • 1.AI Models Fail Intelligence Test
  • 2.Big Tech Loses Addiction Trials
  • 3.Iran War Disrupts Global Markets
  • 4.AI Agent Security Challenges
  • Quick hits on other news
Latest Developments
AI

🧠Every AI Model Scores Under 1% on New Intelligence Benchmark

The Rundown: ARC-AGI-3 launched as a new benchmark testing genuine AI adaptability, where every frontier model scored under 1% while all human testers scored 100%.

The details:

  • Grok 4.2 scored 0% on the benchmark, with other frontier models performing similarly poorly
  • 100% of human testers solved all environments on their first try, highlighting the gap between human and AI reasoning
  • The benchmark tests real-time learning across 135 environments, measuring genuine adaptability rather than pattern matching
  • OpenAI simultaneously shut down Sora, blindsiding Disney and other enterprise partners
Why it matters: This benchmark exposes that current AI models excel at pattern matching but lack genuine reasoning capabilities, suggesting founders building AI products should focus on narrow, well-defined use cases rather than expecting human-level problem solving. The dramatic performance gap indicates we're still years away from artificial general intelligence despite impressive demos.

📰 Source: The Neuron

Share𝕏in
Other

⚖️Meta and YouTube Lose Landmark Social Media Addiction Trials

The Rundown: Meta and YouTube lost groundbreaking California social media addiction trials, ordered to pay $6M total as Section 230 failed to protect them when plaintiffs targeted platform design rather than content.

The details:

  • Meta ordered to pay $4.2M and YouTube $1.8M to a plaintiff who claimed their platforms harmed her mental health as a child
  • Meta separately lost a New Mexico case and was ordered to pay $375M after investigators using decoy minor accounts documented predatory solicitations
  • Over 3,000 similar lawsuits are pending in California, with experts calling this a potential 'Big Tobacco' moment for the industry
  • Both companies plan to appeal, but the precedent suggests Section 230 protections don't extend to algorithmic design choices
Why it matters: These rulings create a new liability framework for platform companies, shifting focus from content moderation to algorithm design and user experience patterns. Founders building social or engagement-driven products should prioritize user wellbeing metrics and design ethical engagement systems to avoid similar legal exposure as regulatory scrutiny intensifies.

📰 Source: Morning Brew

Share𝕏in
Finance

🛢️Iran Conflict Pushes Global Markets Toward Stagflation

The Rundown: The ongoing Iran war has driven oil prices up $1.00 per gallon with the Strait of Hormuz remaining closed, while economic data shows GDP growth slowing to 1% as inflation accelerates back to 4%.

The details:

  • WTI crude surged 37% to $91.84 while Bitcoin and equities trade in inverse correlation to oil prices
  • S&P Global PMI data indicates U.S. economic growth has slowed to just 1% annualized pace
  • Bitcoin miners are operating at $19,000 per coin losses as energy costs surge, accelerating pivots to AI and HPC
  • Iran rejected a U.S. 15-point ceasefire proposal, countering with demands including reparations and sovereignty over the Strait of Hormuz
Why it matters: The combination of slowing growth and rising inflation creates a challenging environment for startups, particularly those with high energy costs like crypto miners and data centers. Founders should prepare for potential capital market tightening and consider how energy price volatility might affect their business models and fundraising timelines.

📰 Source: Milk Road Macro

Share𝕏in
🔥 Synthesized from 8 sources

🔒Supply Chain Attack Compromises Popular AI Library

The Rundown: TeamPCP hacking group compromised the popular LiteLLM PyPI package with 97 million monthly downloads through a sophisticated supply chain attack that installed credential-stealing malware.

The details:

  • Versions 1.82.7 and 1.82.8 were backdoored with Python .pth files that executed malware on every interpreter startup
  • The attack began by exploiting Trivy's GitHub Actions to steal credentials and authentication tokens
  • Developers must immediately pin to version 1.82.6 or earlier and rotate all credentials exposed to compromised environments
  • Google accelerated its post-quantum cryptography migration timeline from 2035 to 2029 citing faster quantum computing advances
Why it matters: This attack demonstrates how AI infrastructure has become a high-value target for sophisticated threat actors, with the potential to compromise thousands of AI applications simultaneously. Founders using AI libraries should implement supply chain security practices, including dependency pinning, security scanning, and credential rotation protocols to protect against similar attacks.

Sources: TLDR +7 others

Share𝕏in

Everything else in the news today

Apple has full access to Google's Gemini model in its own data centers and can distill it into smaller on-device AI models
Reflection AI is raising $2.5B at a $25B valuation with JPMorgan eyeing participation after generating $8B in revenue
Harvey raised $200M at an $11B valuation with ARR nearly doubling from $100M to $190M in five months
Granola raised $125M at a $1.5B valuation for AI-powered meeting assistance
Mastercard agreed to acquire stablecoin infrastructure provider BVNK for up to $1.8B
Circle stock dropped 16% after draft Clarity Act would ban stablecoin yield on USDC balances
Bhutan has sold down its Bitcoin holdings from 13,000 BTC to 4,453 BTC, with outflows exceeding $150M
Apple Maps is introducing ads in the US and Canada this summer as paid search listings
Figma launched beta AI canvas agents that can create and modify design assets using existing design system components
Google's TurboQuant algorithm reduces LLM memory usage by 6x while delivering 8x performance gains
Cisco launched Duo Agentic Identity to monitor autonomous AI agent activity across enterprise workloads
Study shows over 80% of top-ranking pages use AI assistance for content creation
MLB exclusively aired Opening Night on Netflix as part of strategic shift toward new media platforms
SaaStr deployed 30+ AI agents and found only 1 of 5 vendors succeeded by offering Forward Deployed Engineer support before contract signing
Sanders and AOC proposed banning all new data center construction until AI regulation passes
📰TodayFeed📡Signals💰Capital