Copilot drops all Gemini models from web chat; Anthropic projects first quarterly profit while SpaceX S-1 reveals $15B/year compute commitment; White House AI executive order postponed; OpenAI prepares confidential IPO filing.

Run: 72h window (2026-05-19 → 2026-05-22) · 25 items reviewed → 8 published · 6 verified · 2 secondary · 0 rumor · 38% exploration · Run timestamp: 2026-05-22

TL;DR

• GitHub Copilot web drops Gemini — all Gemini models and two GPT variants removed from Copilot Chat on github.com on May 20; OpenAI and Claude models remain. (→ GitHub Copilot web chat drops Gemini models)
• Claude Code agent view and /code-review — May 21 release adds claude agents screen, /code-review with effort levels, and Windows reliability fixes. (→ Claude Code ships /code-review command and agent view)
• Anthropic Q2 profitability claimed, compute costs contested — $10.9B Q2 revenue and $559M operating profit projected; SpaceX S-1 reveals $1.25B/month compute commitment ramping fully after June. (→ Anthropic projects first operating profit as SpaceX S-1 discloses compute deal)
• White House AI executive order postponed — Trump delayed signing a frontier-model review framework on May 21, citing concern it could impede US AI progress. (→ White House AI executive order signing postponed)
• OpenAI prepares confidential IPO filing — Goldman Sachs, Morgan Stanley, and JPMorgan drafting a prospectus targeting September 2026 listing. (→ OpenAI prepares confidential IPO filing)

Items

GitHub Copilot web chat drops all Gemini models and two GPT variants

Source: https://github.blog/changelog/2026-05-20-updates-to-available-models-in-copilot-on-web/ · GitHub Changelog · 2026-05-20 Verification: T2 verified · changelog · dev-tools

GitHub removed all Gemini models and two OpenAI variants — GPT-5.2 Codex and GPT-5.4 nano — from Copilot Chat on github.com on 2026-05-20, citing the goal of consistently ensuring reliable responses. OpenAI and Claude models across price points remain available across all Copilot plans. Users can view the current model picker at github.com/copilot.

Why it matters for automation/productivity: Teams that routed Copilot Chat workflows through Gemini models on the web UI need to switch to an OpenAI or Claude alternative immediately — the change is live. For organizations evaluating multi-model Copilot setups, this narrows the on-web option set to two provider families.

Cross-references:

• https://techsifted.com/posts/github-copilot-changes-may-2026/ (T3, corroborating — broader Copilot plan changes context)

Claude Code ships /code-review command, agent view, and Windows reliability fixes

Source: https://github.com/anthropics/claude-code/releases · Anthropic · 2026-05-21 Verification: T2 verified · changelog · dev-tools Caveats: GitHub release page displayed year 2025 for this version; version sequence (v2.1.147 follows Week 20 versions v2.1.139–142 released May 11–15, 2026) and independent aggregator data confirm 2026-05-21 as the actual date.

Claude Code v2.1.147 released on 2026-05-21 renames /simplify to /code-review with an optional effort level argument (e.g., /code-review high), and fixes auto mode so it no longer suppresses AskUserQuestion when a skill explicitly relies on it. The release also ships reliability improvements: Windows PowerShell failures with winget/Microsoft Store installations are fixed, full-screen strobing in background sessions on Windows Terminal is resolved, and clipboard writes no longer expose copied content in process arguments visible to EDR/SIEM telemetry.

An earlier May release (v2.1.139–142, week of May 11–15) introduced claude agents — a unified screen showing all live Claude Code sessions, their status, and what they are blocked on — along with claude agents --json for scripting integrations with tmux, status bars, and session pickers.

Why it matters for automation/productivity: The /code-review effort-level control and the claude agents session view are directly relevant for teams running multiple parallel Claude Code sessions on large codebases — both reduce the overhead of coordinating multi-agent code tasks.

Cross-references:

• https://releasebot.io/updates/anthropic/claude-code (T3, corroborating — version history)
• https://code.claude.com/docs/en/whats-new (T2, changelog)

Anthropic projects first operating profit ($559M) as SpaceX S-1 discloses the compute deal behind it

Source: https://techcrunch.com/2026/05/20/anthropic-says-its-about-to-have-its-first-profitable-quarter/ · TechCrunch · 2026-05-20 Verification: T2 secondary · financial disclosure via investor briefing · ai-for-business Tier nuance: Primary source is Wall Street Journal (paywalled); TechCrunch and CNBC cite confirmed investor briefing. SpaceX compute-deal figures come from public SEC S-1 filing (T1 for deal terms). COI: Both Anthropic and SpaceX have financial interest in framing the compute arrangement favorably ahead of planned capital raises.

Anthropic informed investors in a funding-round briefing that it projects Q2 2026 revenue of $10.9 billion — a 130% increase from Q1’s $4.8 billion — and its first-ever operating profit of $559 million. Growth drivers cited: enterprise adoption of Claude in coding and cybersecurity workflows, and improved inference cost management. The company cautioned that operating profitability may not hold through the second half of 2026 due to large compute costs scheduled to increase.

On the same day (2026-05-20), SpaceX filed its public S-1 and disclosed the terms of its compute agreement with Anthropic: $1.25 billion per month through May 2029, with payments reduced in May and June 2026 as the arrangement ramps. The deal covers more than 220,000 NVIDIA GB200 GPUs across Colossus 1 and 2, totaling over 300 megawatts of capacity. A 90-day exit clause applies to both parties. An independent analysis at wheresyoured.at argues that the Q2 profit is structurally the result of the promotional May–June pricing window on the compute deal and will not persist when the full $15 billion annual rate applies.

Why it matters for automation/productivity: The financial disclosure affects how organizations should evaluate Anthropic’s long-term stability as a vendor. The $10.9B Q2 revenue run rate signals strong enterprise demand for Claude; the compute-cost trajectory is the offsetting risk. For teams selecting frontier-model providers for multi-year contracts, the H2 compute ramp and upcoming IPO process are the relevant monitoring signals.

Key claims:

• Q2 revenue: $10.9B (+130% from Q1 $4.8B) → WSJ via TechCrunch, investor briefing (T2 secondary)
• Q2 operating profit: $559M → WSJ via CryptoBriefing (T2 secondary)
• Compute deal: $1.25B/month through May 2029 → SpaceX S-1 SEC filing (T1)
• GPU count: 220,000+ NVIDIA GB200s, 300MW+ → Axios reporting on S-1 (T2)
• May–June payments reduced as ramp-up → Axios (T2)

Cross-references:

• https://www.axios.com/2026/05/20/anthropic-spacex-compute (T2, SpaceX S-1 compute terms)
• https://www.wheresyoured.at/anthropics-profitability-swindle/ (T3, adversarial — structural critique of Q2 profit timing)
• https://cryptobriefing.com/anthropic-first-profit-revenue-jump/ (T3, corroborating with caveats on investor-briefing optimism)

White House AI executive order signing postponed by Trump

Source: https://www.washingtonpost.com/technology/2026/05/21/white-house-tore-down-ai-rules-now-its-building-new-defenses/ · Washington Post · 2026-05-21 Verification: T2 verified · news report · policy-regulation

President Trump postponed the signing of a frontier AI executive order on 2026-05-21, hours before the planned ceremony. The order would have established a voluntary 90-day pre-launch review framework for frontier AI models with NSA involvement in classified security testing. Trump told reporters he disliked certain aspects of the order, citing concern that it could impede US AI leadership relative to China. Invitations had already been sent to executives from leading AI companies before the postponement. No revised signing date has been announced.

Why it matters for automation/productivity: The postponement means there is currently no federal pre-launch review requirement for frontier AI models in the US. Organizations building AI governance programs around anticipated federal frameworks should treat those frameworks as still uncertain. Watch for a revised order; the NSA testing component was the element most likely to affect which models can be used in government-adjacent deployments.

Key claims:

• Voluntary 90-day pre-launch review framework with NSA involvement → WaPo (T2)
• Postponed hours before planned ceremony → WaPo, Bloomberg, CNBC (T2, multiple corroborating)
• Invitations sent to tech executives before postponement → Bloomberg (T2)

Cross-references:

• https://www.cnbc.com/2026/05/21/trump-ai-executive-order-postponed.html (T2, corroborating)
• https://www.bloomberg.com/news/articles/2026-05-21/white-house-postpones-ai-cybersecurity-order-signing-by-trump (T2, corroborating)

OpenAI prepares confidential IPO filing targeting September 2026 listing

Source: https://www.axios.com/2026/05/20/openai-ipo-spacex-musk · Axios · 2026-05-20 Verification: T2 secondary · sourced reporting, filing not yet confirmed · ai-for-business

Sources told Axios on 2026-05-20 that OpenAI is preparing a confidential IPO prospectus that could be filed shortly, though timing remains uncertain, with Goldman Sachs, Morgan Stanley, and JPMorgan drafting the document. A public offering is targeted for September 2026. Confidential filings typically precede public S-1 filings by about two months, with offerings following about a month later. OpenAI’s last private valuation was approximately $852 billion; a public offering is targeting up to $1 trillion. CNBC independently confirmed a source saying a filing could come as early as May 22.

Why it matters for automation/productivity: Informational — no immediate workflow leverage. The IPO process will increase transparency around OpenAI’s financial structure. Watch for how API pricing and enterprise terms evolve as OpenAI manages public-market profit expectations, and whether IPO lock-up provisions affect key leadership.

Key claims:

• Goldman Sachs, Morgan Stanley, JPMorgan drafting prospectus → Axios (T2 secondary)
• September listing target → Axios (T2 secondary)
• Last private valuation ~$852B; IPO target up to $1T → CNBC (T2 secondary)

Cross-references:

• https://www.cnbc.com/2026/05/20/openai-ipo-filing.html (T2, corroborating)

Caveats: Filing has not been confirmed as submitted. Timing described as uncertain. Treat as preparatory news, not confirmed action.

Amazon commits $33 billion to AI and cloud infrastructure across Southeast Asia through 2039

Source: https://technode.global/2026/05/21/amazon-plans-over-33b-ai-cloud-infrastructure-investments-in-southeast-asia-by-2039/ · TechNode Global · 2026-05-21 Verification: T2 verified · announcement · ai-for-business

Amazon announced on 2026-05-21 that it plans to invest over $33 billion across Indonesia, Malaysia, Singapore, and Thailand through 2039, focused on AWS cloud and AI infrastructure including data centers. The company projects that these investments will contribute over $64 billion to the collective GDP of those four countries and support more than 56,300 full-time equivalent jobs annually in the data center supply chain. In Indonesia specifically, Amazon signed a 210-megawatt green energy agreement with PT PLN, the state utility, to power the new data centers — the first corporate renewable energy purchase agreement of its kind in the country.

Why it matters for automation/productivity: For organizations in SEA evaluating AWS as an AI infrastructure provider, this signals a long-term regional commitment with physical data center presence. The Indonesia renewable energy deal is a notable first for that market and may affect enterprise sustainability reporting for AWS workloads.

Key claims:

• $33B investment through 2039 across four SEA countries → Amazon/TechNode (T2)
• $64B projected GDP contribution → Amazon (T2)
• 56,300+ FTE jobs annually → Amazon (T2)
• 210MW PT PLN green energy deal (first of its kind in Indonesia) → Amazon/TechNode (T2)

Hark raises $700M Series A at $6B valuation for AI hardware and model platform

Source: https://www.businesswire.com/news/home/20260521171628/en/Hark-Raises-$700M-Series-A-at-a-$6B-Valuation · BusinessWire · 2026-05-21 Verification: T2 verified · press release · ai-for-business

Brett Adcock’s AI startup Hark raised $700 million in a Series A round at a $6 billion post-money valuation on 2026-05-21. Investors include NVIDIA, AMD Ventures, Intel Capital, Qualcomm Ventures, Brookfield, Salesforce Ventures, and lead Parkway Venture Capital. Hark is developing what it describes as a universal interface between humans and machines, combining proprietary AI models and custom hardware designed for speech, vision, and persistent memory. The company, with approximately 70 engineers, plans to grow to 200 and release its first AI models later this summer. No product is currently shipping.

Why it matters for automation/productivity: Informational only — no immediate workflow leverage. Hark’s hardware+model approach is directionally relevant to ambient and persistent AI assistant deployments, but no product yet exists. The investor roster (NVIDIA, Qualcomm, AMD) signals hardware-level differentiation is the core thesis. Revisit when first models ship.

Key claims:

• $700M Series A at $6B valuation → BusinessWire (T2)
• Investors: NVIDIA, AMD Ventures, Intel Capital, Qualcomm Ventures, Brookfield, Salesforce Ventures, Parkway VC → BusinessWire (T2)
• First models planned for later this summer → BusinessWire (T2 — forward-looking, intent not commitment)
• ~70 engineers, growing to ~200 → BusinessWire (T2)

Cross-references:

• https://techcrunch.com/2026/05/21/hark-raises-700m-series-a-for-its-secretive-universal-ai-interface/ (T2, corroborating)

Caveats: No product ships yet. Capabilities have not been publicly demonstrated.

Anthropic expands AI moral formation research, tests ethical-reminder tool in Claude development

Source: https://www.anthropic.com/news/widening-conversation-ai · Anthropic · 2026-05-19 Verification: T2 verified · research announcement · research-papers COI: Anthropic is the author discussing its own alignment practices.

Anthropic published on 2026-05-19 details of an ongoing effort to engage scholars from more than 15 religious, cultural, and secular traditions in shaping Claude’s character and values — what the company calls AI moral formation. During experiments on character development, researchers gave Claude access to an ethical-reminder tool. Internal testing showed markedly lower rates of misaligned behavior on several internal alignment evaluations when Claude used the tool proactively before consequential actions. The announcement previews plans to extend engagement to legal scholars, psychologists, and civic institutions.

Why it matters for automation/productivity: Informational — no immediate product or API change is announced. Longer-term, the ethical-reminder tool is relevant if it becomes a production feature: organizations deploying Claude in high-stakes automated workflows (legal, medical, compliance) would benefit from a built-in behavioral guardrail that activates before consequential steps. Watch for productization signals.

Key claims:

• More than 15 religious and cultural traditions engaged → Anthropic primary (T2)
• Ethical-reminder tool: Claude used it proactively before consequential actions → Anthropic primary (T2)
• Markedly lower rates of misaligned behavior on internal evaluations → Anthropic primary (vendor-claimed; internal evaluation not independently replicated)

Caveats: Evaluation methodology for misaligned-behavior reduction is not disclosed; internal metrics only. Claims should not be treated as independently validated.

Dropped

Items considered but not published, with reason:

Limitations

• Anthropic Q2 profitability primary paywalled: Wall Street Journal broke this story; wsj.com returned a paywall. Figures are cited from TechCrunch and CryptoBriefing, both attributing the WSJ primary. Numbers are treated as T2 secondary, not T2 verified.
• OpenAI.com HTTP 403: openai.com/news/ returned HTTP 403 during this run. OpenAI items were researched via CNBC, Axios, and releasebot.io. No known information gap resulted.
• Claude Code release date uncertainty: GitHub releases page displayed year 2025 for v2.1.147 and v2.1.146; internal version sequence and independent aggregator data confirm 2026-05-21. Item included with caveat in the post.
• xAI primary pages 403: x.ai/news returned HTTP 403. Grok Build details verified via Engadget and The Next Web. Grok Build launch date confirmed as May 14, 2026 (outside window) — item dropped.
• Login-walled coverage: X timelines, Instagram, LinkedIn private feeds, and Discord were not accessed. One dedicated X search pass returned 2 relevant results. Anthropic Colossus II expansion via Daniela Amodei X post surfaced but lacked a blog primary.
• Toto 2.0 window ambiguity: Datadog blog published May 14 (outside window); arXiv paper submitted May 19 (within window). Conservative decision: Datadog primary blog date determines the item window placement — dropped.
• workflow-automation category thin: No new end-user autonomous-agent product launched in the 72h window at GA or broad beta. Google Antigravity 2.0 and Gemini Spark (covered in May 21 bulletin) are the closest recent items.
• mcp-ecosystem category thin: No notable new MCP servers released in window; PulseMCP registry showed most recent additions from May 5–15. Dropped — routine maintenance.
• SEA-region coverage: Dedicated search in Indonesian/SEA terms yielded one in-window item (Amazon $33B investment). No primary product announcements from local AI vendors in the May 19–22 window.
• Geographic and source bias: Run is US/EU-heavy; limited coverage of AI activity in Asia outside Indonesia. Chinese vendor announcements (Alibaba, Baidu, DeepSeek) were not specifically searched.

Search log (compact)

Total searches: 42, of which 16 were exploratory, adversarial, or cross-language (38%).

Suggested next runs

• Claude Code date correction — If Anthropic publishes a primary blog post or the official changelog confirms v2.1.147 as 2026-05-21, upgrade verification from T2 verified with caveat to T2 verified.
• Anthropic H2 compute cost trajectory — The $15B/year SpaceX compute deal becomes fully priced in July 2026; watch for Anthropic investor communications on how this affects the H2 outlook or pricing.
• White House AI executive order revision — No revised signing date announced. Monitor for the revised order text; specifically whether the NSA classified testing component is retained or modified.
• OpenAI IPO prospectus — Confidential filing expected shortly as of May 20. Public S-1 would follow approximately 2 months later. Watch for disclosure of financial details and model pricing strategy under public-market scrutiny.
• Hark first model release — Planned for later this summer. First independent reproduction will determine whether hardware+model integration delivers on the $6B valuation premise.