Rosetta Intel

OpenAI / Anthropic / Google — product and model news; 🛡 marks security-relevant items

OpenAI

Jalapeño inference chip: OpenAI unveils its in-house inference chip, continuing its custom-silicon path to cut inference costs.
GPT-5.6 preview: Previewing under the codenames Sol / Terra / Luna to a small set of government-vetted organizations, with a broader release planned for July.
Sources: ThursdAI· Fortune

Sonnet 5 released: Performance approaching Opus 4.8, introductory pricing of $2/$10 per million tokens (through August 31), able to autonomously use tools such as the browser/terminal.
buffa (Rust protobuf library) DoS 0-day (CVE-2026-55407): Reportedly a crafted malicious protobuf can trigger unbounded heap allocation, causing denial of service. (Single source via the FreeBuf weekly digest; pending direct verification)
Sources: ThursdAI· FreeBuf

NanoBanana 2 Lite: Image-generation model producing images in under 4 seconds, from $0.034 per 1,000 images, with quality above the original NanoBanana.
Gemini 3.5 Pro delayed to July 17: Abandoning the 2.5 Pro architecture for a full rebuild, focused on mathematical reasoning, SVG scene generation, and image quality, positioned against GPT-5.6 and Anthropic Fable 5.
Sources: BigGo Finance· ThursdAI

The most security-relevant signals this issue come from two lines of AI "weaponization": DeepSeek-generated in-browser ransomware, and JADEPUFFER, the fully autonomous LLM ransomware chaining the Langflow RCE (see the AI security section of the main brief for details).
The prompt-injection surface keeps expanding: Mozilla demonstrated indirect prompt injection against AI coding tools via GitHub repositories. Least privilege + sandboxing for agent tool execution remains the primary mitigation.

Source configuration: see intel/sources.yaml