Moonshot says it removed cyber-related data from training phases
Moonshot says it removed cyber-related data from training phases for Kimi K3.1 Pro "to reduce potential misuse risk while preserving general usefulness"
TL;DRMoonshot filtered out hacking data from its AI model to prevent misuse.
Why it matters: Shows AI companies balancing capability with safety—a recurring tension in LLM deployment.
Artificial Analysis says Kimi K3.1 Pro lands 4th on its Intelligence
TL;DRKimi K3.1 Pro ranks 4th globally in AI intelligence benchmarks, strongest open-weights model coming.
OpenAI to Train AI Models Without Internet
TL;DROpenAI is training AI models offline to prevent security breaches during development.
Latest
Three-week on-site audit at OpenAI reveals major gaps in AI safety
OpenAI's safety monitoring misses a quarter of traffic from its most powerful AI models.
Anthropic says its unreleased model is both more capable and better
Anthropic claims a new unreleased model beats Mythos 5.1 in both power and safety alignment.
Instacart Is Exploring a Sale
Grocery delivery app Instacart might get bought after 14 years independent.
SpaceX-Tesla Merger? | Zuckerberg's Deal | The Rundown
Musk explores SpaceX-Tesla merger; Meta secures Zuckerberg; AI agents competition heats up.
Abandoned Popcorn Time apps had to be disabled to prevent a zombie army
Old Popcorn Time piracy app was disabled to stop hackers from hijacking millions of installed copies.
Carta and NewView partner, hoping
Carta and NewView make it easier for startups to let employees sell shares early.
OpenAI Sets Target Valuation of $1.25 Trillion
OpenAI seeks $30B funding at $1.25 trillion valuation.
Global PC shipments fell 21% YoY in Q3
PC sales dropped 21% due to high component costs inflating prices.
From Toast to Bessemer: Who founders will hear from at TechCrunch
TechCrunch's 2026 summit covers practical founder challenges beyond just fundraising and AI.