AI Briefing: May 5 – 7, 2026

Coverage window: May 5 – May 7, 2026 (48 hours)
Generated: 2026-05-07 00:13 UTC
Sources: GitHub API, Twitter/X API, arXiv API, Official Blogs, Web Search, Hacker News

🚨 Breaking (last 24h: May 6–7)

OpenAI GPT-5.5 Instant Goes Live as Default ChatGPT Model

OpenAI rolled out GPT-5.5 Instant as the new default model for all ChatGPT users on May 5, replacing GPT-5.3 Instant. The company emphasizes “significant improvements in factuality” — 52.5% fewer hallucinated claims on high-stakes prompts (medicine, law, finance) and 37.3% fewer inaccurate claims on user-flagged challenging conversations. The model also improves visual reasoning (CharXiv 81.6% vs 75.0%), STEM answers, and web-search routing decisions. OpenAI introduced memory sources to show users exactly which past context informed a personalized response. OpenAI: GPT-5.5 Instant | The New Stack | OpenAI System Card

Anthropic Signs SpaceX Compute Deal, Doubles Claude Code Limits

Anthropic announced a compute partnership with SpaceX on May 6 to use the entire capacity of the Colossus 1 data center — over 300 megawatts and 220,000+ NVIDIA GPUs — available within the month. The deal directly benefits Claude Pro and Max subscribers via doubled Claude Code rate limits, removed peak-hours restrictions, and raised API rate limits for Opus models. Anthropic also disclosed interest in orbital AI compute with SpaceX. The SpaceX deal joins Anthropic’s existing 5 GW Amazon agreement, 5 GW Google/Broadcom deal, $30B Microsoft/NVIDIA Azure partnership, and $50B Fluidstack investment. Anthropic: Higher Limits & SpaceX

OpenAI Unveils MRC Networking Protocol with AMD, Broadcom, Intel, Microsoft, NVIDIA

On May 5, OpenAI published the Multipath Reliable Connection (MRC) protocol — a new open networking standard for AI supercomputer clusters. Already deployed on OpenAI’s NVIDIA GB200 systems at the OCI Abilene site and Microsoft’s Fairwater supercomputers, MRC splits 800Gb/s NICs into multiple parallel planes (e.g., 8×100Gb/s), sprays packets across hundreds of paths, and uses SRv6 source routing instead of BGP for microsecond-scale failure recovery. Partners include AMD, Broadcom, Intel, Microsoft, and NVIDIA. OpenAI: MRC Supercomputer Networking | OCP Spec | Research Paper

GPT-5.5 Matches Claude Mythos in Autonomous Cyberattack Capabilities

The U.K. AI Security Institute (AISI) published findings on May 6 showing GPT-5.5 is the second model ever to complete a 32-step autonomous corporate network intrusion end-to-end, succeeding in 2 of 10 attempts (Claude Mythos Preview achieved 3/10). GPT-5.5 achieved a 71.4% pass rate on the “Expert” tier of offensive cyber tasks, beating Mythos Preview (68.6%). A universal jailbreak bypassed all safety guardrails across malicious cyber queries after six hours of red-teaming; OpenAI updated safeguards but AISI could not fully verify the fix. Decrypt

📊 Market Moves (last 48h: May 5–6)

Anthropic Launches 10 Financial Services Agent Templates

On May 5, Anthropic released 10 ready-to-run agent templates for financial services — pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, GL reconciler, month-end closer, statement auditor, and KYC screener. Each template packages skills, connectors, and subagents. They ship as plugins for Claude Cowork / Claude Code and as cookbooks for Claude Managed Agents. Claude also launched Microsoft 365 add-ins for Excel, PowerPoint, and Word (Outlook coming soon), and expanded data connectors to include Dun & Bradstreet, Fiscal AI, Guidepoint, IBISWorld, SS&C IntraLinks, Third Bridge, Verisk, and a Moody’s MCP app. Claude Opus 4.7 leads Vals AI’s Finance Agent benchmark at 64.37%. Anthropic: Agents for Financial Services

OpenAI Introduces “New Ways to Buy ChatGPT Ads”

OpenAI expanded its advertising offerings on May 5 with new purchasing channels and formats for ChatGPT ads, building on earlier ad-platform announcements. OpenAI: New Ways to Buy ChatGPT Ads

OpenAI ChatGPT Futures: Class of 2026

On May 6, OpenAI announced the inaugural ChatGPT Futures Class of 2026 — 26 students and young builders from 20+ universities (Vanderbilt, UofT, Oxford, Georgia Tech, Waterloo, Smith) receiving $10,000 grants and frontier model access. The program targets the first generation to start and finish college with ChatGPT. OpenAI: ChatGPT Futures

OpenAI Publishes B2B Enterprise AI Report

OpenAI released a “How frontier enterprises are building an AI advantage” report on May 6 analyzing B2B signals and enterprise AI adoption patterns. OpenAI: B2B Signals

🔬 Research (last 7 days)

Most recent arXiv date: May 5, 2026 (20 papers in cs.AI / cs.LG / cs.CL)

Notable Papers

OpenSeeker-v2: Pushing the Limits of Search Agents (May 5) — Improves search agents with informative and high-difficulty query generation. arXiv
Redefining AI Red Teaming in the Agentic Era: From Weeks to Hours (May 5) — Framework for accelerating red-teaming cycles for autonomous agents. arXiv
Rethinking Reasoning-Intensive Retrieval (May 5) — Evaluates and advances retrievers in reasoning-heavy settings. arXiv
SymptomAI: Conversational AI Agent for Everyday Symptom Assessment (May 5) — Agentic healthcare triage system. arXiv
Physics-Grounded Multi-Agent Architecture for Traceable Human-AI Decision Making (May 5) — Multi-agent system with risk-aware decision tracing. arXiv
Logical Consistency as a Bridge: Improving LLM Hallucination Detection (May 5) — Uses label-consistency checking to detect hallucinations. arXiv
Feature-Augmented Transformers for Robust AI-Text Detection (May 5) — Cross-domain and cross-generator AI text detection. arXiv
An Agent-Oriented Pluggable Experience-RAG Skill (May 5) — Experience-driven retrieval-augmented generation for agents. arXiv
From Intent to Execution: Composing Agentic Workflows with Agent Recommendation (May 5) — Automated workflow composition via agent recommendation. arXiv

🛠️ Tools (last 48h: May 5–6)

SDK & Platform Releases

Anthropic SDK Python v0.100.0 (May 6) — Managed Agents multiagents and outcomes, webhooks, vault validation. GitHub
Anthropic SDK Python v0.99.0 (May 5) — Workspace targeting for OIDC federation token exchange. GitHub
OpenAI Python SDK v2.35.0 (May 6) — Image generation API updates. GitHub
OpenAI Python SDK v2.35.1 (May 6) — Fix for imagegen size enum regression. GitHub
OpenClaw v2026.5.4 (May 5) — Google Meet Twilio dial-in via realtime Gemini voice bridge; Windows gateway loopback binding; plugin migration hints. GitHub
OpenClaw v2026.5.5 (May 6) — Feishu topic thread routing fix; LINE DM policy validation; Telegram/Codex progress draft deduplication; xAI Grok 4.3 reasoning effort fix; xAI temperature clamping. GitHub
OpenClaw v2026.5.6 (May 6) — Reverts doctor--fix repair that broke OpenAI Codex OAuth routes; plugin runtime fetch header normalization; debug proxy header capture fix. GitHub

Hacker News Trending (May 6)

Cloudflare Agents can now create accounts, buy domains, and deploy — Agent-native infrastructure expansion. HN
Telus Uses AI to Alter Call-Agent Accents — Real-time voice modification in customer service. HN
Vibe coding and agentic engineering are getting closer than I’d like — Simon Willison on the tension between agentic automation and engineering rigor. HN
Show HN: Tilde.run – Agent sandbox with transactional, versioned filesystem HN
Wiki Builder: Skill to Build LLM Knowledge Bases — DAIR Academy on Claude Code plugin for wiki construction. HN

💭 Industry Pulse (last 48h: May 5–6)

Key Voices on X/Twitter

@sama (Sam Altman): “ChatGPT feels very ‘switched on’ now” (May 6). Celebrated GPT-5.5 Instant rollout: “the new instant model in chatgpt is so good damn” and “5.5 in codex is so good for non-coding tasks” (May 5). Twitter
@gdb (Greg Brockman): Announced “codex is for everyone” (May 6). Promoted the MRC protocol as “a new open networking protocol for large AI training clusters, deployed in production” (May 6). Noted “Major ChatGPT upgrade rolling out now, in the form of GPT-5.5 Instant” (May 5). Twitter
@AndrewYNg: “Coding agents are accelerating different types of software work to different degrees. When we architect teams, understanding these differences matters” (May 5). Twitter
@ylecun (Yann LeCun): Political commentary dominated his feed: “RFKjr == Lyssenko” and “Trump’s war on science” (May 4–6). Twitter
@JeffDean: Announced keynote at CAISconf; celebrated Google Translate’s 20th anniversary (April 28). No AI-research tweets in the 48h window. Twitter
@karpathy: No new tweets in the 48h window; most recent was April 30 (Sequoia Ascent fireside chat). Twitter

Emerging Discussions

Anthropic “Dreaming” for Managed Agents: Twitter reports (May 7) indicate Anthropic is testing a scheduled “dreaming” process where Claude reviews recent work and updates memory autonomously — effectively agentic self-improvement during downtime. Twitter
Claude + SpaceX reactions: Multiple X posts noted the irony of Elon Musk’s SpaceX powering a key OpenAI competitor, with some joking about “Project kill open ai” and asking whether Anthropic will now get “mythos” compute. Twitter

🌄 Sources & Data Provenance

Source	Status	Notes
GitHub API	✅ Working	9 releases retrieved from Anthropic, OpenAI, OpenClaw
Twitter/X API (twitterapi.io)	✅ Working	8 queries executed; individual accounts reliable
arXiv API	✅ Working	20 papers retrieved from May 5; most recent date 2026-05-05
Web Extract (Anthropic blog)	✅ Working	2 official posts extracted (SpaceX, Finance Agents)
Web Extract (OpenAI blog)	✅ Working	5 official posts indexed; 3 detailed extracts obtained
Web Search (Firecrawl)	✅ Working	5 results retrieved for GPT-5.5 coverage
Hacker News	✅ Working	May 6 front page extracted; May 5 rate-limited
Wiki Raw Archive	✅ Available	May 1–4 ingestions present and queryable

Every claim above links directly to its primary source. Unlinked observations were cross-referenced from multiple sources.

AI Briefing: 2026-05-07