Friday 26 June 2026

The Brief – 26/06/2026

The top things worth knowing about in AI today.

  1. OpenAI limits GPT-5.6 to trusted partners

    OpenAI released a preview of GPT-5.6 on 26 June, three models named Sol, Terra and Luna, but is restricting access to a small group of vetted partners at the White House's request while the government reviews their capabilities. Wider availability is expected within weeks. OpenAI said it does not want pre-release government sign-off to become the default, warning it would keep the best tools from the developers, enterprises and cyber defenders who need them.

    Read more
  2. OpenAI and Broadcom unveil first inference chip

    OpenAI and Broadcom revealed Jalapeno, OpenAI's first custom processor built specifically to run large language models. The two companies took it from initial design to manufacturing tape-out in nine months, aiming to lower the cost of serving models at scale. It is a step toward OpenAI owning more of its own hardware rather than relying entirely on Nvidia.

    Read more
  3. Meta building a prediction market app

    Meta is building a standalone app, codenamed Arena, where users guess the outcome of real-world events, putting it up against Kalshi and Polymarket. Its Llama model will write the questions from trending topics and decide the yes-or-no result in near real time, with no human review. Kalshi and Polymarket have already cleared 130 billion dollars in trading volume this year; Meta plans to start with points rather than real money.

    Read more
  4. AI takes 80 percent of venture funding

    Global venture investment reached 330 billion dollars in the first quarter of 2026, and AI companies absorbed more than 80 percent of it. Four rounds alone, OpenAI, Anthropic, xAI and Waymo, accounted for 188 billion dollars. The numbers show capital concentrating in a handful of large labs rather than spreading across the wider startup market.

    Read more
  5. AWS adds tools to keep AI agents reliable

    At its New York summit, AWS expanded Bedrock AgentCore with tools to monitor, test and secure AI agents running in production. New features turn live agent traces into fixes, run A/B tests on agent changes, and screen every agent action for prompt injection and data leaks. The emphasis on reliability reflects a wider shift from agent demos to systems businesses can actually depend on.

    Read more
  6. KPMG and Microsoft deploy AI agents globally

    KPMG and Microsoft expanded their partnership to roll out AI agents across KPMG's global operations using Microsoft Agent 365 and Copilot. The deal puts governance, identity and audit controls around the agents so they can be managed and monitored at scale. It is a sign that large professional services firms are moving agents from pilots into day-to-day client work.

    Read more
  7. Google delays Gemini 3.5 Pro to July

    Google has pushed the launch of Gemini 3.5 Pro to July, saying it needs more time to refine how the model handles complex tasks. Its smaller sibling, Gemini 3.5 Flash, is already available and now powers the default AI Mode in Search. The delay is a reminder that even the leading labs are pacing releases around reliability rather than rushing every model out.

    Read more
The Weekly AI Brief

Practical AI, distilled.

A short read every week — the few things worth your time, and nothing that isn’t.