AI news daily brief: 2026-05-22
Seven stories from May 21-22, 2026: one leadership crisis, two regulation updates, three product launches, and one model rollout. The day's biggest story, a separate standalone deep-dive on a frontier model announcement, runs at a separate URL.
GitHub fights for survival inside Microsoft as leadership vacuum and competitor pressure mount
The Verge reported on May 21, 2026, citing current and former GitHub employees, that GitHub is fighting for its survival amid a wave of outages, security incidents, and competitive pressure from Cursor and Claude Code. GitHub has operated without a CEO since Thomas Dohmke resigned in summer 2025; the remaining leadership reports to Microsoft's CoreAI team under Jay Parikh, who sources described as poorly regarded inside Microsoft. At least 11 former GitHub employees have since joined Dohmke's new developer platform startup Entire, per The Verge's count of Entire's listed staff. The Information reported separately that Parikh privately warned colleagues GitHub "faces a critical threat." Microsoft also reportedly considered acquiring Cursor to close Copilot's gap with rivals, and The Verge reported earlier in the same week that Microsoft cancelled many of its internal Claude Code licenses to pressure its own engineers to improve GitHub Copilot instead. Read more: The Verge.
OpenAI launches ChatGPT add-in for Microsoft PowerPoint
OpenAI launched a ChatGPT add-in for Microsoft PowerPoint on May 21, 2026. Users install it from the Microsoft Office Marketplace by searching for ChatGPT under Home > Add-ins and signing in with an OpenAI account. Per OpenAI's product page, the add-in reads the deck structure and lets users create slides from notes, rewrite content, restructure narrative, add sections, and summarize presentations. OpenAI said the add-in preserves editable slide content and keeps users in review before sharing. The add-in follows an earlier ChatGPT for Excel add-in, per The Verge. No standalone pricing page was found at time of writing; the product page links to the Microsoft Office Marketplace for installation. Read more: ChatGPT for PowerPoint.
Trump pulls planned AI executive order hours before signing after Sacks, Zuckerberg, and Musk pushed back
President Trump cancelled the signing of a planned AI and cybersecurity executive order on May 21, 2026, hours before the event was to take place. Per Axios, the order would have required frontier AI developers to share new models with the US government 90 days before public release and established a Treasury-led cybersecurity clearinghouse for AI companies. Trump told reporters: "I didn't like certain aspects of it. I postponed it. I think it gets in the way of, you know, we're leading China, we're leading everybody, and I didn't want to do anything to get in the way of that lead." Axios reported that White House AI adviser David Sacks "hated it" and that critics inside the White House called the order "something doomers wanted." Meta CEO Mark Zuckerberg, xAI CEO Elon Musk, and Sacks all spoke with Trump between Wednesday night and Thursday morning before the signing was cancelled, per Axios sources. The Washington Post reported on May 22, 2026, that last-minute lobbying by tech industry officials drove the decision. Read more: Axios.
Proton Pass adds AI access tokens for credential sharing with agents
Proton Pass launched AI access tokens on May 21, 2026, adding a credential-sharing layer for AI agents. Per the Proton blog, users create a token that grants an agent read-only access to specific vaults (not all credentials) and can set an expiration ranging from one hour to one year. Every time an agent uses a credential, an audit log entry is created that includes the reason the agent gave for the access. Setup requires copying a setup instruction block from Proton Pass settings and pasting it to the AI agent. Per Proton, the feature ships at no extra cost on Pass Plus (included in Proton Unlimited), Pass Family, Pass Professional, and Proton Workspace. Proton positioned the feature as an alternative to the common workaround of creating a throwaway account with a weak password for agents to use. Read more: Proton blog.
GitHub Copilot ships semantic issue search and task-aware model routing in two May 20 updates
GitHub shipped two generally available features on May 20, 2026. The first, semantic issue search, lets developers ask Copilot Chat on GitHub.com in plain language to find issues by intent rather than by exact keyword. Per the GitHub changelog, the feature uses a semantic issues index and surfaces issues that are related even when worded differently. That helps developers who cannot remember an issue title or want to filter by environment. The second, auto model selection in VS Code, routes Copilot requests to the model best suited to the task, weighing real-time availability and reliability signals alongside task dimensions including reasoning complexity, code generation difficulty, and bug diagnosis. Per the changelog, users can hover a response to see which model was used, switch between Auto and a specific model at any time, and auto respects admin-set model policies. Both features are live on all Copilot subscription tiers. Read more: GitHub changelog: semantic search, GitHub changelog: auto model selection.
xAI adds Grok to OpenCode terminal agent for SuperGrok and X Premium subscribers
xAI published instructions on May 21, 2026, for connecting a Grok account to OpenCode, an open-source terminal-based coding agent. Per xAI's blog post, users install OpenCode via the official install script and then run the /connect command inside the tool to authenticate with xAI. Two authentication methods are available: xAI Grok OAuth for SuperGrok subscribers, which opens a browser for sign-in, and a headless token flow for X Premium subscribers who work over SSH or on remote hosts. Once connected, the agent runs Grok Build, the same model that powers xAI's own terminal coding agent CLI. Per xAI, more open-source agent integrations are planned. Read more: xAI blog.
Gemini 3.5 Flash becomes Google's default model across 900 million monthly active users
Google confirmed at Google I/O 2026 on May 19, 2026, that Gemini 3.5 Flash is now the default model in the Gemini app and AI Mode in Google Search, reaching 900 million monthly active users, up from 400 million one year earlier, per Google's announcement. Per the Google blog post, Gemini 3.5 Flash outperforms Gemini 3.1 Pro on coding and agentic benchmarks, including a Terminal-Bench 2.1 score of 76.2%, while running at Flash-tier speed. The model is available for developers in the Gemini API (AI Studio and Android Studio) and via the Gemini Enterprise Agent Platform. Google has not confirmed a public launch date for Gemini 3.5 Pro. Read more: Google blog.
Sources
- GitHub faces a fight for its survival at Microsoft: The Verge, named current and former employees, May 21 2026
- Microsoft executives sound alarm on GitHub's eroding AI lead: The Information, secondary
- ChatGPT for PowerPoint: OpenAI product page
- ChatGPT for PowerPoint generates presentations with prompts: The Verge, secondary
- Anti-"doomer" feedback derails Trump's AI executive order: Axios, named sources, May 21 2026
- Last-minute lobbying by tech industry officials led Trump to cancel AI order: Washington Post, secondary
- Proton Pass: A password manager for AI agents: Proton official blog, May 21 2026
- Semantic issue search in Copilot Chat: GitHub changelog, May 20 2026
- Auto model selection now routes based on your task in VS Code: GitHub changelog, May 20 2026
- Use Grok in OpenCode: xAI official blog, May 21 2026
- Gemini 3.5: frontier intelligence with action: Google blog, May 19 2026