May 12, 2026 — curated links and takeaways.
1. Google says hackers used AI to develop a major security flaw
Cybercriminals used AI to discover and develop a zero-day exploit, marking the first documented case of AI-assisted vulnerability discovery in the wild. Security teams need to reassess threat models assuming adversaries now have AI-powered reconnaissance capabilities.
2. Google's Gemma 4 release on May 4, 2026 marks significant moment for open-source community
Google released the Gemma 4 family on May 4, 2026, bringing open-weight reasoning models to developers. This democratizes frontier reasoning capabilities previously locked behind API paywalls, reshaping the open-source competitive landscape.
3. Claude Mythos Preview available to 11 organizations for security vulnerability detection
Anthropic launched Claude Mythos (Project Glasswing) preview on April 7 to select enterprises for cybersecurity applications, signaling specialized model variants targeting high-value verticals. No general availability planned, positioning this as premium offering.
4. OpenAI raises $110 billion at $730 billion valuation
OpenAI closed a $110 billion funding round in February 2026 led by Amazon ($50B), SoftBank ($30B), and Nvidia ($30B), setting record for private tech fundraises. Capital inflow signals sustained belief in frontier model scaling despite ongoing safety debates.
5. Sarvam AI open-sources Sarvam 30B and 105B models with mixture-of-experts architecture
Sarvam AI released Sarvam 30B and 105B (with 10.3B active parameters via MoE) in February 2026, adding competitive open-weight options for inference efficiency. MoE architecture at scale represents emerging approach to balancing capability and cost in open models.