AI security: Anthropic Claude Mythos/Glasswing limited rollout/OpenAI $122B/Stargate threats
Key Questions
What is Anthropic doing with its Claude Mythos model?
Anthropic limited rollout of Claude Mythos, a powerful cyber model capable of zero-days, due to cyberattack risks. It excels at identifying vulnerabilities but is too dangerous for public release. Status is 'climaxing', first seen 2026-03-13, last updated 2026-04-16.
What is Project Glasswing?
Anthropic built Project Glasswing to limit access to its most powerful AI cyber model to key partners for patches. It addresses fears of hackers using the model for cyberattacks. This sweeping cybersecurity initiative restricts Mythos preview.
What major funding did OpenAI secure?
OpenAI closed a $122 billion funding round at $852B post-money valuation. This supports projects like Stargate in UAE amid elevated cyber risks. NA VC reached $252B.
How is Japan supporting AI development?
Japan relaxed privacy laws to become the easiest country for AI development. Microsoft is betting $10B in Japan for AI pivot. This aids AI infra amid global competition.
What AI deals involve Google and Meta?
Google finalized a $32B deal to acquire cybersecurity startup Wiz for cloud/AI security. Meta will open-source new models and build a superintelligence hardware team. NVDA involved in open-source AI evolution.
What are the cyber risks highlighted in AI advancements?
Anthropic fears Mythos could enable future cyberattacks, leading to access limits. Cyber risks elevated with models reaching advanced coding levels. Anthropic's measures aim to mitigate hacker exploitation.
What visions does OpenAI have for the AI economy?
OpenAI envisions public wealth funds, robot taxes, and a four-day work week. This ties into its massive funding and Stargate threats. Broader AI infra includes UAE projects.
How is open-source AI evolving?
Open-source AI evolves rapidly, with Meta open-sourcing models, NVDA refining training. Leaked OpenAI GPT-Image-2 shows advanced capabilities. X rolls out Grok-powered features like translation.
Anthropic limits Claude Mythos cyber model (zero-days) via Glasswing to key partners for patches; OpenAI $122B/Stargate UAE; Meta/Google Wiz/NVDA; Japan relaxes privacy for AI dev; NA VC $252B; cyber risks elevated. (first seen: 2026-03-13, last updated: 2026-04-16)