📰

subtl daily briefing

Share𝕏in
AllAICryptoSaaSFinanceStartups

Good morning, tech leaders. The industry just witnessed what could be its 'Big Tobacco moment' — Meta and YouTube lost their first major social media addiction trial in California, with Section 230 protections failing to shield them. Meanwhile, AI is having an identity crisis as every frontier model flunked a new intelligence test that all humans passed with flying colors.

In today's briefing

  • 1.Big Tech Loses Addiction Trial
  • 2.AI Models Fail New IQ Test
  • 3.Supply Chain Attacks Hit LiteLLM
  • ⚡Quick hits on other news
Latest Developments
Other

⚖️Meta and YouTube Lose Historic Social Media Addiction Trial

The Rundown: Meta and YouTube were found negligent in a landmark California social media addiction trial and ordered to pay $6M total in damages, marking the first time Section 230 protections failed against platform design claims.

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
  • ●Section 230 protections failed because plaintiffs targeted platform design rather than content moderation decisions
  • ●Over 3,000 similar lawsuits are pending in California, with Meta separately ordered to pay $375M in a New Mexico child safety case
Why it matters: This ruling fundamentally shifts liability for social platforms from content to design decisions, potentially forcing expensive algorithmic changes and opening floodgates for thousands of pending lawsuits. For founders, this signals the end of the 'move fast and break things' era — platform design decisions now carry real legal consequences.

📰 Source: Multiple sources

Share𝕏in
AI

🧠Every AI Model Fails New Intelligence Test That Humans Ace

The Rundown: ARC-AGI-3, a new benchmark testing genuine AI adaptability, revealed that every frontier model scored under 1% while 100% of human testers solved all problems on their first try.

The details:

  • ●Grok 4.2 scored 0% on the test, with all other frontier models including GPT-4 and Claude performing similarly poorly
  • ●The benchmark tests genuine problem-solving adaptability rather than pattern recognition from training data
  • ●OpenAI simultaneously shut down Sora and blindsided Disney partnership, while Sanders and AOC proposed banning new data center construction until AI regulation passes
Why it matters: This exposes a critical gap between AI hype and actual intelligence — current models excel at pattern matching but fail at true reasoning. For AI startups, this suggests we're still far from AGI and highlights the importance of building AI tools for specific, well-defined tasks rather than general intelligence applications.

📰 Source: The Neuron

Share𝕏in
🔥 Synthesized from 8 sources

🚨Major Supply Chain Attack Compromises LiteLLM Package

The Rundown: TeamPCP hacking group compromised the popular LiteLLM PyPI package (97M monthly downloads) through a sophisticated supply chain attack, installing credential-stealing malware via Python .pth files.

The details:

  • ●Attackers backdoored versions 1.82.7 and 1.82.8 with malware that executed on every Python interpreter startup to steal credentials and tokens
  • ●The attack chain began by exploiting Trivy's GitHub Actions to gain access to the LiteLLM repository and PyPI publishing credentials
  • ●Developers must immediately pin to version 1.82.6 or earlier and rotate all credentials that may have been exposed
Why it matters: This attack demonstrates how a single compromised dependency can expose millions of AI applications to credential theft. For technical founders, it underscores the critical importance of dependency pinning, supply chain security audits, and having incident response plans for when your core infrastructure gets compromised.

Sources: TLDR +7 others

Share𝕏in
⚡

Everything else in the news today

Google accelerated post-quantum cryptography migration deadline to 2029, six years ahead of schedule due to quantum advances→
Reflection AI is raising $2.5B at a $25B valuation with JPMorgan participation after generating $8B in revenue→
Harvey raised $200M at $11B valuation with ARR nearly doubling from $100M to $190M in 5 months→
Granola raised $125M at $1.5B valuation for AI meeting notes→
Apple has full access to Google's Gemini model in its own data centers for distilling smaller on-device AI models→
Bitcoin miners are losing ~$19,000 per block as production costs hit $88,000/BTC against ~$69,200 spot price→
Circle stock dropped 16% after Clarity Act draft would ban stablecoin yield on USDC balances→
Mastercard agreed to acquire stablecoin infrastructure provider BVNK for up to $1.8B→
Iran rejected US 15-point ceasefire proposal, countering with demands including reparations and Strait of Hormuz sovereignty→
Google's TurboQuant compression reduces LLM memory usage by 6x while delivering 8x performance gains→
SES AI CEO says most Western battery companies are dying, pivoting firm to AI-powered materials discovery→
Figma launched beta AI canvas agents that generate design assets using existing design system components→
Apple Maps introducing ads in US and Canada this summer, ending privacy-first advantage over Google Maps→
Halter raised $220M for AI-powered cow collar platform tracking livestock health and breeding cycles→
FCC banned new routers manufactured outside the US on security grounds, affecting most major consumer brands→
CERN successfully transported antimatter by truck for the first time→