The Brief · Issue 1
Week of 29 June 2026 — the EU softens its AI Act as Anthropic ships Claude Sonnet 5
The EU Council waters down AI Act deadlines, Anthropic launches Claude Sonnet 5, and Databricks previews sub-second SQL.
Compiled on 22 September 2026 for The Brief's launch archive, from sources published between 29 June and 5 July 2026.
Four stories that mattered
The EU Council gives AI Act deadlines more room
On 29 June 2026 the Council of the European Union gave final approval to the Digital Omnibus on AI, a package of amendments to the EU AI Act (Regulation (EU) 2024/1689) that the European Parliament had already approved on 16 June. The Omnibus pushes back the compliance deadline for high-risk AI systems to 2 December 2027 for standalone systems and 2 August 2028 for those embedded in regulated products, and gives systems already on the market until 2 December 2026 to meet the Article 50 requirement to watermark AI-generated audio, image, video and text. It also adds an outright ban, from the same date, on AI systems built to generate non-consensual intimate imagery or child sexual abuse material.
Teams building or buying tools that classify, score or generate content should re-check which of their systems fall into the high-risk bucket under the new timeline — the deadline moved, but the obligation to classify has not.
Source: Licentium, EU Council Formally Adopts AI Omnibus Amending EU AI Act, 29 June 2026
Anthropic launches Claude Sonnet 5
Anthropic released Claude Sonnet 5 on 30 June 2026, describing it as its most agentic Sonnet model yet, with gains in tool use, coding and long-horizon reasoning that bring it close to its flagship Opus tier at a fraction of the cost. The model launched at $2 per million input tokens and $10 per million output tokens — introductory pricing Anthropic later made permanent.
For analytics teams, the relevant change is agentic reliability: a cheaper model capable of extended tool-calling and terminal use lowers the cost of building pipelines that call an LLM repeatedly inside a workflow, from data-cleaning agents to automated report generation.
Source: Anthropic, Introducing Claude Sonnet 5
Databricks previews sub-second SQL on the lakehouse
On 30 June 2026 Databricks put Lakehouse Real-Time into public beta: a new serverless SQL warehouse type built for sub-second reads against Unity Catalog tables at hundreds to thousands of concurrent users. It targets a gap most lakehouse platforms have left to specialised real-time databases — serving live dashboards and operational applications straight off governed data without a separate serving layer.
Teams currently exporting a slice of the lakehouse into a dedicated real-time store for latency-sensitive dashboards may want to test the beta before that duplicate pipeline gets built again.
Source: Databricks, June 2026 release notes
Tool moves
- Databricks added Claude Sonnet 5 to Model Serving the same day Anthropic launched it, giving customers Foundation Model API access to the model for coding and agentic workflows. Source: Databricks, June 2026 release notes
- The SQL alert task for Databricks Lakeflow Jobs reached general availability, letting teams branch downstream jobs on an alert's result. Source: Databricks, June 2026 release notes
- Parquet v2 reached general availability on Databricks. Source: Databricks, June 2026 release notes
- LLM governance through Databricks model services entered beta. Source: Databricks, June 2026 release notes
One how-to
Build a five-minute AI system inventory. With the EU AI Act's deadlines now fixed rather than open questions, the cheapest thing a data or analytics team can do this week is list what it already has, before a compliance team asks.
- List every internal tool that scores, classifies, ranks or generates content about a person — credit models, churn scores, resume screens, content generators — not just the obvious "AI products."
- For each one, note who built it (vendor or in-house), what it decides or influences, and whether that decision has a legal or similarly significant effect on someone.
- Flag anything touching employment, credit, essential services, law enforcement or biometric identification — the categories the AI Act treats as high-risk.
- Record which AI agents act autonomously rather than just producing a recommendation for a human to approve; autonomous action raises the governance bar.
- Note the model or vendor behind each system, so a retirement or price change (see Tool moves, most weeks) doesn't surprise you.
This isn't a full responsible AI programme — it's the inventory that any real data governance or EU AI Act compliance effort has to start from, and most teams don't have one written down.
One number
53.3% — the ISM Manufacturing PMI for June 2026, released 1 July 2026, down 0.7 point from May but still marking a sixth straight month of expansion (any reading above roughly 47.5% over time indicates growth). The Employment sub-index, by contrast, stayed in contraction territory at 49.7%.
What it doesn't tell you: the PMI is a diffusion index built from purchasing managers' answers to "better, same, or worse," not a measure of actual output levels — it can rise or fall sharply on shifting sentiment even when production barely moves, and it covers manufacturing only, not the much larger services sector.