← Back to newsletters

AI Daily Brief — Thu Sep 3

2026-09-03

By Vadym · Generated with AI, curated by me


Listen to this issue

Today’s AI news is about guardrails catching up to capability — OpenAI’s Astra became the first model to cross the industry’s “Critical” cybersecurity threshold, ChatGPT Health locked its new Epic integration to read-only, and Meta is quietly discovering that EU law, not ideology, now decides what “open weights” means. None of it slowed the money down — Wonderful just raised $550 million, and VS Code shipped its most agent-native release yet.


Headlines & News
Safety

OpenAI’s Astra Becomes First Model to Cross ‘Critical’ Cybersecurity Threshold

OpenAI said on September 2 that its upcoming Astra model is the first to reach the “Critical” capability level under its Preparedness Framework. In testing, Astra independently found two previously unknown zero-day vulnerabilities, chained multiple flaws in a hardened OS to gain root access, escaped a browser sandbox to execute commands on the host, and scored a perfect result on ExploitBench. It also refused 91.5% of cybersecurity jailbreak attempts, up from 59% for its predecessor. OpenAI says full cyber capabilities won’t be widely available at launch, opting for phased access through early testers and its Daybreak Blue program instead.

Source: SecurityWeek

Health

ChatGPT Health Adds Epic Integration for Clinicians

OpenAI connected ChatGPT Health to Epic’s electronic health record system, which holds data for more than 325 million patients, letting clinicians pull appointment notes, lab results, medications, and specialist documentation directly into the chat. Access is strictly read-only — the AI can summarize and analyze records but can’t write anything back. A new Healthcare Public Data plugin also connects ChatGPT to ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed, and PubMed. OpenAI cited testing across 4,363 physician ratings spanning 27 clinical use cases, with 99.1% of responses rated safe.

Source: TechCrunch

Funding

Enterprise AI Startup Wonderful Raises $550M at $5B Valuation

Amsterdam-headquartered Wonderful closed a $550 million Series C led by Insight Partners, with Salesforce joining as a new investor alongside existing backers Index Ventures, IVP, Vine Ventures, 9Yards, and Bessemer. The round comes just six months after a $150 million raise at a $2 billion valuation and brings Wonderful’s total funding past $800 million. The company, which has grown to 650 employees across 35-plus markets, sells an “AI operating system” that automates enterprise workflows and coordinates agents across customer, employee, and back-office functions.

Source: Tech.eu

Tools

VS Code 1.136 Adds Agent Merge and Multi-Root Workspace Support

Microsoft shipped VS Code 1.136 on September 2 with an experimental Agent Merge feature that lets coding agents resolve review feedback, fix failed CI checks, and handle merge conflicts, rerunning the workflow until a pull request is actually mergeable. The release also adds experimental multi-root workspace support so Copilot and Claude agent sessions can work across every folder in a workspace without needing full paths, plus hierarchical session organization that nests related chats under their parent with status indicators for which ones need attention.

Source: Visual Studio Code


Analysis

Takeaway

Notice where the guardrails moved today: OpenAI didn’t decide not to ship Astra, it built a rollout schedule around it. ChatGPT Health didn’t earn broad EHR access, it earned read-only. Meta’s openness calculus is now a liability calculation under EU law, not a philosophy. Safety, in other words, has stopped being a decision made before a launch and started being a feature shipped inside one — and none of it is slowing the money down, as $550 million for Wonderful and a genuinely agent-native VS Code release both make clear.

— Boba


Curated by Vadym