Sunday 13 September 2026

The Brief – 13/09/2026

The top things worth knowing about in AI today.

  1. Anthropic CEO calls for slowing AI progress

    Dario Amodei published a 3,800-word essay, "We Must Pace the Frontier", arguing that model capabilities are improving faster than researchers can verify the systems stay under control. He is not calling for a training freeze: the proposal is to slow the rate of capability gains so safeguards catch up, and Anthropic has committed to giving third-party evaluators permanent, employee-level access to its systems. A day earlier Sam Altman told OpenAI staff the company would consider pacing its own development if peer labs did the same.

    Read more
  2. AI agent swarm breached 395 organisations

    GreyNoise documented an attacker running hundreds of AI agents built on OpenAI's Codex harness and a DeepSeek model against PaperCut print servers, compromising at least 440 instances across 395 organisations in 48 countries. The campaign went from an empty workspace to remote code execution on a live target in under four hours, and hit 11 organisations in 26 seconds once it scaled. Credentials were harvested from 280 victims and 12 lost domain admin, with schools making up roughly half the targets. Unpatched edge software is now being probed at machine speed.

    Read more
  3. OpenAI asks Congress for binding safety rules

    On 9 September OpenAI called for mandatory federal AI safety requirements that trigger once a system crosses defined capability thresholds, covering testing standards, third-party evaluation, cybersecurity and incident reporting. It also backed four California bills and said it will keep supporting state legislation until Congress acts, a reversal of the industry's earlier push for federal pre-emption. Congress adjourns in December.

    Read more
  4. Mistral raises 3 billion euro on sovereignty demand

    The French lab closed a 3 billion euro Series D led by Samsung Electronics, with EQT's Scaleup Europe Fund and PSG Equity as co-leads, at a post-money valuation above 21 billion euro. The money goes to compute, infrastructure and international expansion, including 1GW of European capacity by 2030. Its pitch is that customers choose which region processes their queries, the same data-sovereignty argument Australian government and regulated-industry buyers keep raising with US vendors.

    Read more
  5. Australian AI inquiry closes submissions 14 September

    The federal Joint Select Committee on Artificial Intelligence, appointed on 20 August, closes submissions on 14 September and must report by 30 November. Its terms of reference specifically name the barriers small, medium and family businesses face in adopting AI, making it one of the few openings for operators rather than vendors to put evidence on the record. It runs alongside the Office of AI in Prime Minister and Cabinet, with Australian Standards for AI expected to reach legislation in early 2027.

    Read more
  6. OpenAI opens its Codex agent infrastructure

    The Agents API entered public beta on 10 September, giving developers the same harness that runs Codex: long-lived sessions, MCP and custom tools, subagent coordination, context compaction, and execution in an OpenAI-hosted sandbox or your own compute through partners including Cloudflare, Vercel, Oracle and DigitalOcean. There is no separate API fee, with usage billed through the models and tools each session consumes. For most teams it removes the reason to build an agent orchestration layer in-house.

    Read more
  7. Pentagon weighs $5 billion loan to Fluidstack

    The Wall Street Journal reports the Pentagon's Office of Strategic Capital is in talks to lend roughly $5 billion to AI cloud startup Fluidstack, directed at domestic supply and manufacturing capacity for data centre components rather than a single facility. It would be the office's largest loan by a wide margin. Nothing is signed, but it marks defence budgets moving into commercial AI infrastructure financing.

    Read more
  8. Gemini desktop app arrives on Windows

    Google released the Gemini app for Windows 10 and 11 globally on 10 September. Alt+Space opens it over whatever is on screen for summaries, fact-checks and image generation, and it can pull from Gmail and Drive or hand longer jobs to the Gemini Spark agent. Spark is not in the Windows build yet, running only on mobile, Mac and web.

    Read more
  9. Microsoft merges its two Copilot apps

    The unified Copilot app, a single client handling both personal and work accounts, begins its worldwide Windows and Mac rollout in mid-September after mobile and web shipped in August. It keeps the Microsoft Copilot name with a new icon and is the first structural step toward the single AI app Microsoft has said it will ship later this year. IT teams should note that consumer Deep Research, Podcasts and group chat threads were retired on 18 August and do not carry across.

    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.