AI Frontier · Jun 26, 2026

Product, model, and talent developments at OpenAI / Anthropic / Google DeepMind. 🛡 = security-related.

OpenAI

• OpenAI × Broadcom inference chip(6/25) — the two jointly announced an inference chip optimized for LLMs, advancing OpenAI's custom-silicon strategy with the aim of reducing dependence on third-party GPUs and driving down inference costs.
  CNBC
• OpenAI has previously stated publicly that prompt injection is a "frontier safety challenge" with no clean solution — especially salient as MCP and agents roll out at scale.

Anthropic

• Continued poaching from Google— Gemini contributors Jonas Adler(AI coding) and Alexander Pritzel(model training) are leaving Google for Anthropic. Earlier, Nobel laureate John Jumperjoined Anthropic and star researcher Noam Shazeerjoined OpenAI. With both companies nearing IPOs, pre-IPO equity has become a key draw.
  Bloomberg· TechCrunch

Google DeepMind / AI

• Gemini 3.5 Pro still unreleased— Pichai previously promised "give us another month"; June was supposed to be the key launch window but has not delivered.
• The steady loss of senior researchers (to Anthropic / OpenAI) has raised investor doubts about Google's competitive pace.
  TechCrunch

Also Worth Watching

• MiniMax M2.5 released— positioned as a low-cost alternative to frontier models; June saw a dense wave of Chinese vendor releases approaching frontier performance at markedly lower API cost.
  llm-stats

---

Configuration: see intel/sources.yaml · ai_frontier section