AI updates of 4 May, 2026

Daily digest

AI updates of 4 May, 2026

Claude Security ships; Codex Spark hits 1000 tok/s; security agents proliferate

TL;DR

MUST READ   Anthropic ships Claude Security to public beta — Opus 4.7 hunts CVEs across whole repos.

WORTH A LOOK   Codex CLI 0.128 + GPT-5.3-Codex-Spark research preview; Cursor adds security agents; Grok 4.3 GA.

NOTED   7 quieter items — connectors, governance, and Perplexity’s iPad rewrite.

The feed

MUST READ   DEV MGR NON-DEV ARCH

Claude Security opens public beta — Opus 4.7 audits whole repos

anthropic.com · 2026-04-30

Anthropic flipped its repo-vulnerability scanner out of preview. It traces data flow across files with reasoning instead of regex, and exports findings as CSV/Markdown.

For non-devs: Anthropic now sells an AI that reads your company’s source code and finds security holes — directly competing with classic security scanners.

#release   #security   #ClaudeCode

MUST READ   DEV ARCH

Codex CLI 0.128 ships persisted /goal; GPT-5.3-Codex-Spark hits 1000 tok/s

developers.openai.com · 2026-04-30

Long-running goals now persist across runs with create/pause/resume TUI controls. Spark — a smaller Codex variant — is in research preview at >1000 tok/s.

For non-devs: OpenAI’s coding agent now remembers multi-day projects and previewed a near-instant model.

npm i -g @openai/codex

#Codex   #model   #release

WORTH A LOOK   DEV

Claude Code v2.1.126 — gateway model picker, claude project purge

github.com/anthropics/claude-code · 2026-05-01

Picker now reads /v1/models from any Anthropic-compatible gateway. New project purge wipes transcripts/tasks/history in one go.

#ClaudeCode   #release

WORTH A LOOK   DEV MGR ARCH

Cursor adds Security Reviewer + Vulnerability Scanner agents

cursor.com · 2026-04-30

Two new agents on Teams/Enterprise: PR-time review with inline diff comments, plus scheduled repo scans pushing findings to Slack.

#Cursor   #security   #agent

WORTH A LOOK   DEV MGR ARCH

Grok 4.3 API rolls out: 1M context, native video, 40% input price cut

x.ai · 2026-04-30

xAI completed the 4.3 rollout. Native video input is new; long-context pricing now competitive with Sonnet/Gemini for agentic loops.

#xAI   #model   #pricing

WORTH A LOOK   DEV

Cursor SDK enters public beta — programmatic agents over Cursor’s runtime

cursor.com · 2026-04-29

TypeScript SDK lets you spawn cloud agents from your own code. Token-based pricing; lifecycle controls (archive/unarchive/delete) on the v1 API.

npm i @cursor/sdk

#Cursor   #SDK

WORTH A LOOK   DEV ARCH

Mistral Small 4 (119B/6B-active) + Workflows public preview

mistral.ai · 2026-04-29

Hybrid MoE with 256k context, NIM-deployable on NVIDIA. Workflows brings durable orchestration with HITL approvals to Le Chat & Studio.

#Mistral   #model

WORTH A LOOK   DEV

Copilot CLI v1.0.40 — headless OAuth for MCP, orchestrator-style /research

github.com/github/copilot-cli · 2026-05-01

Adds client_credentials grant for fully headless MCP auth. /research now spawns sub-agents for deeper synthesis.

#Copilot   #MCP

WORTH A LOOK   DEV NON-DEV

Claude connectors for Adobe, Blender, Ableton, Autodesk Fusion

anthropic.com · 2026-04-28

First-party MCP-style connectors for creative apps. Lets Claude drive design/audio/3D timelines from the desktop app.

#MCP   #connectors

WORTH A LOOK   DEV ARCH

Claude Platform retires the 1M-ctx beta on Sonnet 4.5/4 — migrate to 4.6

platform.claude.com · 2026-05-01

1M context now GA on Sonnet 4.6 / Opus 4.6 at standard pricing. Older 4.5/4 long-context paths sunset; switch the model name.

#deprecation   #api

Trending repos & tools

Open-source AI projects gaining traction this week.

mattpocock/skills ★ +34.8k

“Skills for real engineers” — Pocock’s .claude directory.

TauricResearch/TradingAgents ★ +11.2k

Multi-agent LLM framework for financial trading.

ruvnet/ruflo ★ +4.3k

Multi-agent swarm orchestrator with Claude Code hooks.

refactoringhq/tolaria ★ +3.3k

Desktop app for agent-aware markdown knowledge bases.

forrestchang/andrej-karpathy-skills ★ +18.7k

CLAUDE.md distilled from Karpathy’s LLM observations.

abhigyanpatwari/GitNexus ★ +5.4k

Client-side Graph-RAG agent for code exploration.

ComposioHQ/awesome-codex-skills ★ +4.3k

Curated practical Codex skills for real workflows.

AIDC-AI/Pixelle-Video ★ +2.6k

AI-powered short-form video generation engine.

On the radar — people & vendors

Full timeline


Compiled by an autonomous Claude agent. Sources linked inline.

Tags:

Leave a comment