OpenAI and Broadcom on Wednesday unveiled Jalapeño, the first custom AI inference chip in a long-term partnership. The ASIC ships to data centers by end of 2026 and targets large language model inference at scale.

The 13-person startup says its custom-trained models hit 10x to 100x token efficiency on enterprise knowledge. Notion, Harvey, and Microsoft are already piloting it.

Coval raised $28M led by Norwest with Base10, Twilio Ventures, and Y Combinator, bringing total funding to $31M. The platform simulates and monitors AI voice agents so enterprises can ship them without silent failures.

Seltz, rebuilding web search for AI agents, raised $12.5M in seed funding from Speedinvest and B Capital. The startup owns its own crawler, index, and ranking and pitches itself as a retrieval layer for LLM agents.

Parts of the NSA have lost access to Anthropic's Mythos 5 after a Commerce Department export-control order on June 12, 2026, with the dispute now blocking access for Five Eyes allies and for non-U.S. staff at Anthropic.

Maxima's new agent, Max, prepares journal entries, reconciliations, and flux analysis for accountants to review, with hard human-in-the-loop gates and SOX-ready audit trails. Early customers report 80% less prep time.

Snyk's Evo Agentic Development Security governs what AI coding agents use, do, and produce, with controls inside the agent execution loop. Snyk's own data shows 1 in 12 MCP-server installs carries a critical finding.

Anthropic's Claude Code lead Boris Cherny told Meta's @Scale that agents prompting agents is the next era of coding, and that long-running recursive loops will define how teams ship software for the rest of 2026.

NVIDIA used DTW Ignite 2026 to pull SoftBank, AdaptKey, Amdocs, NTT DATA, ServiceNow, TCS, Forsk, VIAVI, and KDDI into one telecom agent stack built on synthetic data, NemoClaw, OpenShell, and digital-twin simulation.