AI Radar

AI Radar — 31 May 2026

4 items 3 verified 1 secondary 0 rumor 12 sources 34% exploration

Lighter signal window after the dense May 28 Anthropic cluster: OpenAI adds Computer Use on Windows to Codex, Microsoft ships A2A GA in Copilot Studio, DeepSeek locks in its 75% price cut, and the EU’s bid to access Anthropic’s Mythos stalls on US government objection.

Run: 26–31 May 2026 (5-day expanded; strict 72h window May 28–31 yielded 2 items not covered in prior bulletins) · 22 items reviewed → 4 published · 3 verified · 1 secondary · 0 rumor · 34% exploration · Run timestamp: 2026-05-31


TL;DR


Items

OpenAI Codex adds Computer Use on Windows and mobile initiation

Source: https://developers.openai.com/codex/changelog · OpenAI Developers · 2026-05-29 Verification: T2 verified · announcement · workflow-automation / dev-tools

OpenAI released Codex version 26.527 on May 29, 2026, adding Computer Use support on Windows. The capability allows Codex to see, click, and type within Windows desktop applications in the foreground while executing a task — the same interaction mode previously available on macOS. Remote initiation from iOS or Android via the ChatGPT mobile app is also now supported, allowing users to start or monitor Codex tasks on a Windows machine from a phone. On Windows, Computer Use runs on the active desktop and takes foreground control during execution; it cannot operate in the background while a concurrent Windows session is in use. Computer Use on Windows is not available in the European Economic Area, the United Kingdom, or Switzerland at launch.

Why it matters for automation/productivity: Developers and QA engineers on Windows can now direct Codex to autonomously navigate and operate any Windows application — executing test runs, filing bug reports, or reviewing app output — without requiring API access to the target application. The mobile initiation pathway extends these autonomous coding workflows to scenarios where the developer is away from the machine.

Key claims:

Cross-references:

Caveats: Foreground-only on Windows — execution takes over the desktop and cannot run alongside a concurrent user session. EEA/UK/Switzerland restriction in effect at launch; no timeline for regional expansion announced. Codex access requires eligible plan; not available to all ChatGPT tiers.


Microsoft Copilot Studio May 2026 update ships A2A communication GA and voice agents GA

Source: https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/new-and-improved-computer-using-agents-a-new-workflows-experience-and-real-time-voice-experiences/ · Microsoft Copilot Blog · 2026-05-26 Verification: T2 verified · announcement · workflow-automation

Microsoft published the Copilot Studio May 2026 update on May 26, confirming several generally-available capabilities. Agent-to-agent (A2A) communication is now generally available, allowing Copilot Studio agents to delegate tasks and coordinate work across systems without human handoff at each step. Real-time voice agents reached general availability in North America through Dynamics 365 Contact Center, with support for caller identification, question answering, and context-preserving handoffs between AI and human agents. An early-release redesigned workflows canvas introduces agent nodes that embed AI execution within deterministic workflow steps. Microsoft also notes an improved orchestrator in early release achieving approximately 20% higher evaluation performance and 50% lower net token consumption compared to the prior version. Computer-using agents — enabling agents to operate website and desktop UIs directly without API support — had reached general availability separately on May 13, 2026.

Why it matters for automation/productivity: A2A GA in Copilot Studio removes the need for custom orchestration code between agents in enterprise automation stacks built on Power Platform. Multi-agent flows — for example, a research agent delegating to an approval agent delegating to a data-entry agent — can now be assembled within Copilot Studio without external middleware.

Key claims:

Cross-references:

Caveats: Orchestrator performance figures are vendor-measured. Voice agents GA restricted to North America at launch. A2A communication requires Power Platform licensing; not available in all SKUs. Redesigned workflows experience is early release, not GA.


DeepSeek makes V4-Pro 75% price cut permanent at May 31 deadline

Source: https://api-docs.deepseek.com/quick_start/pricing · DeepSeek · effective 2026-05-31 (announced ~2026-05-22) Verification: T2 verified · pricing announcement · model-release

DeepSeek’s 75% promotional discount on V4-Pro API pricing became permanent on May 31, 2026, when the original promotion window was scheduled to close. The new list price is $0.435 per million input tokens and $0.87 per million output tokens — compared with the pre-promotion list prices of $1.74/M input and $3.48/M output. Cache-hit input tokens are priced at $0.00363/M. DeepSeek announced the permanence of the discount approximately May 22, 2026, ahead of the effective date. No changes were made to model capabilities or API availability alongside the pricing update.

Why it matters for automation/productivity: The permanent pricing locks DeepSeek V4-Pro as a cost-efficient option at scale for high-volume API workloads. For teams evaluating API cost baselines across providers — batch document processing, automated code review, long-horizon agent runs — the pricing floor established today is a stable planning input.

Key claims:

Cross-references:

Caveats: Announcement date (~May 22) falls outside the 5-day expanded window; item is included because the effective date of the pricing change is May 31, 2026. API pricing is subject to future revision; no contractual lock-in implied. DeepSeek is a Chinese-origin service — data residency and terms of service should be reviewed before use in regulated workflows.


EU access to Anthropic Mythos stalled as US blocks non-US sharing

Source: https://www.cnbc.com/2026/05/29/mythos-ai-models-eu-talks-us.html · CNBC · 2026-05-29 Verification: T3 secondary · policy reporting · policy-regulation Tier nuance: CNBC primary returned HTTP 403 from this environment. Claims corroborated via The Star (T3), Blockonomi (T3), CDT Europe AI Bulletin (T3), and the Insurance Journal (T3).

The European Commission sought access to Anthropic’s Claude Mythos Preview for security evaluation and testing, according to a May 29 CNBC report citing European Commission spokesperson Thomas Regnier. Anthropic told the Commission that the White House would need to authorize sharing Mythos with non-US entities, and the US administration opposed such access due to security concerns. Commission officials planned direct meetings with Anthropic to pursue access arrangements. In parallel, Anthropic stated on May 28 that it expects to bring Mythos-class models to all customers “in the coming weeks,” which the Commission views as a potential alternative pathway. EU AI Office enforcement powers begin in August 2026; at least one Commission official cited that timeline as a fallback mechanism for compelling access.

Why it matters for automation/productivity: Informational only for most commercial teams — no immediate workflow leverage. For organizations in the EU conducting AI risk assessments, AI Act compliance reviews, or procuring AI tools for security-relevant workflows, the standoff signals that access to frontier cybersecurity-capable models may remain jurisdictionally restricted through at least Q3 2026, independent of whether those models reach general commercial availability.

Key claims:

Cross-references:

Caveats: All sourcing for this item is T3 (primary CNBC unavailable). Anthropic’s Mythos wide-release announcement (see 2026-05-29 bulletin) may render the bilateral access dispute moot once general rollout proceeds — but no GA date has been committed. EU-US dynamics may shift ahead of August 2026.


Dropped

Items considered but not published, with reason.

Title consideredSourceReason
Anthropic Mythos wide-release announcementanthropic.com/news/claude-opus-4-8 (2026-05-28)Covered in 2026-05-29 bulletin
OpenAI Rosalind Biodefense programopenai.com (2026-05-29)Covered in 2026-05-30 bulletin
OpenAI Frontier Governance Frameworkopenai.com (2026-05-28)Covered in 2026-05-30 bulletin
Google Pay MCP server public previewdevelopers.googleblog.com (2026-05-27)Covered in 2026-05-29 bulletin
GPT-5.5 Instant response style updateopenai.com (2026-05-30)Incremental quality improvement to existing model; no new capability shipped
Codex Pro 2× promotion expirydevelopers.openai.com (2026-05-31)Minor pricing footnote; subsumed into Codex Computer Use item context
Project Glasswing initial update (10K+ bugs)anthropic.com/research/glasswing-initial-update (2026-05-22)Outside 5-day expanded window (primary dated May 22)
Indonesia launches .ai.id AI-specific domainPANDI/KORIKA (2026-05-29)SEA-region exploration find; low BD-actionability for public-digest audience
Microsoft Build 2026 pre-announcementsMultiple (conference June 2–3, 2026)Forward-looking only; no products shipped as of May 31
Claude for Small Business launchanthropic.com (2026-05-13)Outside window (May 13)
Anthropic Finance Agent templatesanthropic.com (2026-05-05)Outside window (May 5)
Adobe AI Productivity Agentnews.adobe.com (2026-05-06)Outside window (May 6)
OpenAI ChatGPT for Excel/Google Sheets GAopenai.com (2026-05-05)Outside window (May 5)
OpenAI GPT-5.3 Instant Mini fallback modelopenai.com (April 2026)Outside window
NSA MCP security advisorynsa.gov (2026-05-20)Outside window (May 20)
GitHub trending repos (Understand-Anything, etc.)github.com/trendingStars-only discovery; no primary product launch announcement found in window
Hugging Face Daily Papers (May 28–30)arxiv.org (2026-05-28 to 2026-05-30)T3 preprints; abstracts did not surface deployment implications within this window
OpenAI GPT-5.6 reference in Codex logsMultiple (undated)T5 — undated, screenshot-sourced, no verifiable primary

Limitations


Search log (compact)

Q: "Anthropic Claude AI announcement May 31 2026" → 9 results, 2 high-relevance
Q: "OpenAI ChatGPT new release announcement May 31 2026" → 9 results, 3 high-relevance
Q: "Google DeepMind Gemini AI release May 2026" → 9 results, 2 high-relevance
Q: "xAI Grok new release announcement May 2026" → 8 results, 3 high-relevance
Q: "new AI model release launched May 29 30 31 2026" → 9 results, 1 high-relevance
Q: fetch anthropic.com/news/finance-agents → May 5 item, outside window
Q: fetch anthropic.com/news/claude-for-small-business → May 13 item, outside window
Q: "OpenAI GPT-5.3 Instant Mini ChatGPT release May 31 2026" → 9 results, 0 new in-window
Q: "ChatGPT Excel Google Sheets integration launch May 2026" → 10 results, 0 new in-window (May 5 GA already covered)
Q: "Anthropic Mythos AI model announcement May 2026" → 7 results, 3 high-relevance
Q: fetch red.anthropic.com/2026/mythos-preview/ → April 7 primary, prior coverage
Q: "MCP Model Context Protocol new servers announcement May 29 30 31 2026" → 10 results, 0 new in-window
Q: "AI agent framework launch release May 29 30 31 2026 LangChain AutoGen CrewAI" → 10 results, 0 new in-window
Q: "AI dev tools Cursor Claude Code update release May 29 30 31 2026" → 9 results, 0 new in-window
Q: fetch anthropic.com/glasswing → Project Glasswing details
Q: "EU Mythos AI cybersecurity 'talks' 'allied' May 29 2026 CNBC" [lateral exploration] → 9 results, 6 high-relevance ✓
Q: "site:x.com AI launch OR release May 30 31 2026" [social exploration] → 8 results, 0 new in-window
Q: "AI announcement May 30 OR May 31 2026 site:techcrunch.com OR site:theverge.com" → 8 results, 0 in-window
Q: fetch anthropic.com/news → index page, no May 29-31 items beyond May 28
Q: fetch cnbc.com/2026/05/29/mythos-ai-models-eu-talks-us.html → 403 Forbidden
Q: "AI Indonesia startup AI Asia Tenggara Mei 2026" [SEA mandatory exploration] → 9 results, .ai.id domain discovered
Q: "site:github.com topic:mcp-server stars AI agent 2026" [GitHub exploration] → 8 results, no in-window primary launch
Q: fetch radiostar.harianjogja.com/2026/05/30/ → .ai.id domain details confirmed (May 29-30)
Q: "GitHub MCP security scanning AI agents May 2026" → 9 results, GA date May 5 (outside window)
Q: "OpenAI news announcement May 29 30 31 2026 new feature product" → 9 results, 4 high-relevance (Codex Windows found) ✓
Q: "Anthropic Mythos wide release blog post May 28 29 2026" → 7 results, confirmed in 2026-05-29 bulletin
Q: fetch anthropic.com/research/glasswing-initial-update → May 22 primary, outside expanded window
Q: "OpenAI Codex 'Computer Use' Windows expand May 30 2026" → 9 results, 7 high-relevance ✓
Q: fetch bloomberg.com/news/2026-05-28/anthropic → paywalled, Mythos in prior bulletin
Q: "Anthropic Mythos wide release criticism safety concerns May 2026" [adversarial] → 8 results, Pentagon controversy context found
Q: "new AI product launch announcement productivity agent framework May 29 30 2026" → 9 results, 2 high-relevance
Q: fetch developers.openai.com/codex/changelog → May 29 Computer Use on Windows confirmed ✓
Q: fetch news.adobe.com/news/2026/05/ → May 6, outside window
Q: "EU Mythos AI models US government talks policy regulation cybersecurity May 2026" → 9 results, 5 high-relevance
Q: "Microsoft Build 2026 announcements Windows AI agents Copilot preview May 30 31" [exploratory] → 9 results, conference June 2-3
Q: "AI product release tool announcement May 30 2026 new feature" → 8 results, 1 high-relevance
Q: "DeepSeek V4 pricing permanent discount May 31 2026" [exploratory] → 9 results, 7 high-relevance ✓
Q: "Codex Pro promotion expires May 31 2026 OpenAI pricing change" → 6 results, minor footnote
Q: fetch microsoft.com/copilot-studio May 2026 blog → CUA/A2A/voice GA details confirmed ✓
Q: "DeepSeek V4 Pro pricing permanent announcement date official May 2026" → 9 results, May 22 announcement confirmed
Q: "Copilot Studio 'computer-using agents' generally available May 2026 date" → 9 results, May 13 CUA GA confirmed
Q: fetch buildfastwithai.com/blogs/ai-news-today-may-30-2026 → Glasswing, Build 2026 context; no new items
Q: fetch techcommunity.microsoft.com Copilot Studio CUA GA → certificate error
Q: "OpenAI Codex Computer Use Windows adversarial criticism limitations May 2026" [adversarial] → 8 results, EEA/UK/CH restriction confirmed

Total searches: 46, of which ~16 exploratory, adversarial, or SEA/lateral (34%). Primary source fetches: 9 (anthropic.com/news, red.anthropic.com, anthropic.com/glasswing, anthropic.com/research/glasswing-initial-update, developers.openai.com/codex/changelog, microsoft.com/copilot-studio, cnbc.com 403, radiostar.harianjogja.com, bloomberg.com paywalled).


Suggested next runs