Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡

Here’s what we got for ya today:

Let’s get right into it!

Sign Up | Advertise | Tools

The creator of Claude Co-Work says AI is already reshaping how software gets built, shifting engineers from writing code line-by-line to managing fleets of AI agents that generate, test, and iterate on their behalf. That change, he argues, is less about automation and more about leverage, freeing developers to focus on product decisions and user needs. Co-Work itself grew out of an unexpected signal: users began pushing a coding tool into non-coding workflows, from analytics to operations, revealing demand for a more general-purpose AI collaborator.

The bigger shift, though, is organizational. Companies seeing the strongest results aren’t just adopting AI tools, they’re rebuilding workflows around them, driving productivity gains that far exceed incremental improvements. While still early, adoption is accelerating quickly as compute scales and models improve. In the near term, AI is removing routine work; longer term, he compares the moment to the printing press, with coding becoming a baseline skill and a surge of new startups emerging as barriers to building collapse.

→ Watch video on YouTube

The White House is drafting an executive order that would require AI companies to submit new frontier models to a formal government vetting process before they reach the public. National Economic Council Director Kevin Hassett told Fox Business the review would mirror how the FDA evaluates drugs — models would need to be demonstrated safe before launch rather than releasing them and patching problems after deployment.

The push follows last week's controversy over Anthropic's Mythos model, which can autonomously identify network vulnerabilities at a level that alarmed officials. The proposed order would give the government early access to frontier models, raising real questions about whether a Washington review queue will meaningfully slow down the current pace of AI releases — or create a regulatory bottleneck that hands advantage to foreign labs operating without the same constraints.

→ Read more at Tom's Hardware

OpenAI quietly rolled out a more permissive version of GPT-5.5 — code-named "Spud" — to a select group of vetted cybersecurity researchers on Wednesday. The release gives defenders access to a model that security testing shows is nearly as capable as Anthropic's Mythos at identifying and exploiting software vulnerabilities, a capability that had previously been locked behind strict access controls.

The move represents a deliberate shift in how AI labs are treating dual-use capabilities. Rather than keeping dangerous offensive functionality off the table entirely, OpenAI is channeling it toward authorized researchers — a bet that arming the defenders first shifts the cyber balance before the models leak into the wrong hands.

→ Read more at Axios

Elon Musk's xAI launched Grok Connectors on Wednesday, enabling users to link Grok directly to Google Workspace (Gmail, Drive, Docs, Sheets, Calendar), Notion, GitHub, SharePoint, and OneDrive. The integrations mean users can now manage emails, update slides, read calendar events, and analyze spreadsheets without switching out of Grok — turning what was a standalone chatbot into a workflow control center.

The launch puts Grok in direct competition with ChatGPT's plugin ecosystem and positions xAI as a serious enterprise player, not just a consumer AI. For professionals juggling five or more apps daily, this is the most concrete pitch yet that a single AI interface can replace the constant context-switching — assuming users trust Grok with full access to their Google accounts.

→ Read more at xAI