Tech brief
Anthropic Eyes $7B Decart Deal as OpenAI Unlocks Offensive Cyber AI
Frontier AI developers double down on Israeli infrastructure optimization and specialized security research models.
Tech
Anthropic is acquiring Decart to control critical compute optimization layers, cementing Israel's role as a core hub for frontier AI hardware architecture.
BackgroundDecart was founded in late 2023 by Unit 8200 veterans and builds specialized software that optimizes AI model training and real-time inference speed across hardware chips. Global tech giants are competing fiercely for infrastructure software capable of cutting massive AI compute costs.
- Anthropic is finalizing a deal valued between $6 billion and $7 billion to acquire Tel Aviv-based AI infrastructure startup Decart.
- Decart's platform drastically improves real-time inference efficiency across Nvidia, AWS, and Google processors used for frontier LLM workloads.
- The acquisition would integrate roughly 100 deep-tech engineers into Anthropic and establish its primary international R&D operations center in Israel.
- The transaction underscores how specialized Israeli deep-tech talent is commanding premier valuations during the global AI compute arms race.
Tech
OpenAI is giving defensive security researchers offensive AI tools, arming cybersecurity teams while raising the stakes for access control.
BackgroundFrontier AI developers have faced pressure to balance strict safety guardrails against the practical demands of cybersecurity teams who need realistic exploit simulation tools. Identity-gated access frameworks allow platforms to restrict dangerous offensive capabilities to verified defensive researchers.
- OpenAI unveiled GPT-5.6-Cyber, achieving a 95% completion rate on authorized vulnerability discovery and exploit validation tasks.
- The model is released under the identity-gated Daybreak framework, separating public models from high-capability cyber research tools.
- Israeli cyber defense firms and security testing startups gain less-restricted access to test zero-day defensive frameworks against automated exploits.
- Industry experts warn that permissive offensive capabilities require strict identity verification to prevent unauthorized misuse by state-backed threat actors.
Unlock the full brief
Sign in to read every signal, takeaway, and source. Free account — Apple, Google, or email.
