Rosetta Intel

Daily updates from OpenAI / Anthropic / Google (🛡 = security-related)

OpenAI

Weighing a 2027 IPO— OpenAI now expects rival Anthropic to go public first; both have made confidential filings with the U.S. SEC. OpenAI had at one point targeted an autumn listing.
Bloomberg

Claude Tag on Slack (beta)— teams can @Claude in a Slack channel to delegate tasks, letting it build up context incrementally, work asynchronously, and operate with controlled accessto available tools and data. Currently in beta for Claude Enterprise & Team customers. The controlled tool/data access is the notable security design point.
Anthropic Newsroom
Reportedly weighing an IPO as soon as October; the confidential filing was confirmed in early June (TechCrunch).
TechCrunch

Gemini 3 + Gemini 3 Deep Think— DeepMind calls Gemini 3 its "best multimodal understanding and best agentic / coding model"; Gemini 3 Pro is in preview in the Gemini app, AI Studio, and Vertex AI, with Deep Think as an enhanced reasoning mode.
Google Blog
Managed Agents (Gemini API, public preview)— lets developers build and deploy autonomous, stateful agents running inside securely isolated, Google-hosted Linux sandboxes. Worth watching from an agent security and isolation standpoint.
Google Cloud
Gemini 3.5 Flash GA— the GA release aimed at agentic / coding tasks with sustained frontier performance.

All three vendors are productizing "agent + sandbox + controlled tool access" as a main line (Anthropic's Claude Tag controlled access, Google's Managed Agents isolated sandbox) — directly related to the "lethal trifecta" risks emphasized in this week's OWASP Agentic AI security report.

Configuration: see the ai_frontier category in intel/sources.yaml