ma2tic

The brief · 7 April 2026

The AI brief, 2026-04-07

EY is putting all 160,000 of its audits through an AI agent, NIST is drafting its playbook for critical infrastructure, and Anthropic is lending its best hacker out to 50 handpicked companies. Four stories, four ways of gauging just how deep AI has already wormed its way into the machinery that matters.

2 min read J / K to navigate

EY rolls out AI agents across all 160,000 audits worldwide

If your company gets audited by EY, your next review will partly run through AI agents instead of juniors combing through invoices one by one. EY is targeting full automation by 2028, though how that plays out at this scale remains to be seen.

practitioners › Multi-agent framework built with Microsoft, integrated into EY Canvas, announced April 7, 2026 as part of a multi-billion-dollar investment.

EY →

SandboxAQ Open-Sources Dataset to Speed Up Solid-State Battery Discovery

Finding new solid-state battery materials usually takes years of computationally expensive simulations. AQVolt26 provides a dataset plus machine-learning-trained interatomic potentials designed to simulate these materials much faster. It has nothing to do with everyday ChatGPT use, but it shows how generative AI is also pushing forward materials science research.

practitioners › AQVolt26 pairs a specialized dataset with a suite of MLIPs built for solid-state batteries, released by SandboxAQ.

SandboxAQ →

Anthropic hands its best hacking model to 50 hand-picked companies

Claude Mythos Preview finds and exploits software vulnerabilities better than any of Anthropic's public models. Instead of a public release, Anthropic is giving it to AWS, Apple, Microsoft, Google, Cisco and roughly 40 others to harden their code before attackers get there first. This changes how software giants stress-test their own systems, not yet how you use Claude day to day.

practitioners › Project Glasswing, announced April 7, 2026: gated access to Claude Mythos Preview for ~50 partners, $100M in committed credits.

Anthropic →

NIST Drafts AI Guidance for US Critical Infrastructure

This won't touch you directly if you're just using ChatGPT day to day. But if your water, power, or telecom network ever runs on an AI-managed system, frameworks like this are what decide the safeguards working behind the scenes.

practitioners › Concept note published April 7, 2026, a profile derived from NIST's AI RMF, aimed at critical infrastructure operators.

NIST →