AI Radar — 16 Jul 2026

Grok Build open-sourced after researcher proves silent whole-repo upload; EU emergency antitrust order restores ChatGPT across WhatsApp for 500M Europeans; Anthropic ships self-serve HIPAA and Admin API Beta; 16 Nobel Laureates join 200+ economists warning on AI economic pace; Fable 5 free window extended for third time.

Run: Jul 13–16 2026 (strict 72h window) · 30 reviewed → 5 published · 3 verified · 2 secondary · 0 rumor · 65% exploration

Note: Items from the Jul 12–15 window (Claude for Teachers, China companion-AI enforcement, White House Gold Eagle, OpenAI Codex Micro, Anthropic Canada $10M CAD) were published in the Jul 15 bulletin and are not re-published here.

TL;DR

• Grok Build open-sourced — xAI released Grok Build’s full Rust source on July 15 after researcher cereblab proved v0.2.93 silently uploaded entire git repositories (5.10 GiB, vs 192 KB the model needed) to a Google Cloud Storage bucket, including .env credentials. (→ Grok Build)
• ChatGPT back on WhatsApp EEA — EU Commission issued a rare Article 8(1) emergency antitrust order forcing Meta to restore free ChatGPT access on WhatsApp for 30 EEA countries; access resumed July 13 for roughly 500 million users. (→ ChatGPT WhatsApp)
• Anthropic HIPAA self-serve — Healthcare and API orgs can now sign the BAA, download the implementation guide, and enable HIPAA configuration without a sales call; Admin API Beta adds programmatic user management for Enterprise. (→ Anthropic Enterprise)
• Nobel economists on AI — 16 Nobel Laureates joined 200+ economists on July 13 calling for urgent institutional action; notable signatories include Daron Acemoglu and Simon Johnson, who previously pushed back on AI displacement projections. (→ Nobel statement)
• Fable 5 window extended — Anthropic extended free Fable 5 access to July 19 for the third time; after that date the model costs $10/$50 per million input/output tokens. (→ Fable 5)

Items

Grok Build Open-Sourced After Silent Whole-Repo Upload Incident

Source: https://github.com/xai-org/grok-build · xAI · 2026-07-15 Verification: T2-T3 secondary · open-source release + security incident · dev-tools

Security researcher cereblab published wire-level analysis on July 12 showing that Grok Build CLI v0.2.93 packaged and transmitted entire git repositories — full commit history and all — to a Google Cloud Storage bucket named grok-code-session-traces, regardless of which files the agent actually accessed. On a 12 GB test repository, model-turn traffic used 192 KB while the storage channel moved 5.10 GiB — a roughly 27,800× gap proving whole-repo transmission independent of agent activity. Files the agent was explicitly instructed not to read were still uploaded; .env credentials appeared verbatim in both the model-turn and storage channels. xAI applied a server-side flag (disable_codebase_upload: true) on July 13 to stop uploads, and committed to deleting all previously uploaded user data. On July 15, xAI released the full Rust source at github.com/xai-org/grok-build — an architectural fix that allows local operation and removes xAI from the data pipeline entirely.

Why it matters for automation/productivity: Any development team that used Grok Build CLI before July 13 should audit whether credentials, .env secrets, or proprietary code were included in repositories uploaded to the grok-code-session-traces GCS bucket. For teams evaluating AI coding agents, the incident sets a concrete benchmark question: does the tool confine telemetry to files the model actually reads, and can that be verified in source? The open-source release enables auditing but does not recover data uploaded before July 13.

Key claims:

• Grok Build v0.2.93 uploads entire repos to grok-code-session-traces (GCS); never-read canary file recovered from git bundle → gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547 (T2, 2026-07-12)
• 192 KB model-needed vs 5.10 GiB uploaded on a 12 GB repo → cereblab gist (T2, 2026-07-12)
• xAI server-side fix applied July 13; full Rust source open-sourced July 15 → github.com/xai-org/grok-build (T2, 2026-07-15); cryptobriefing.com (T3, 2026-07-15)
• Elon Musk: “zero anything whatsoever will remain” on user data → neowin.net citing Musk X post (T3, 2026-07-14)

Caveats: x.ai/news returned HTTP 403 during this run; open-source release confirmed via GitHub repo and multiple T2-T3 trade outlets. Musk’s data deletion pledge is T2 for the attribution only — not confirmed-deletion. cereblab’s analysis covers v0.2.93; behavior of earlier versions not established. Transmission to cloud ≠ use in model training — xAI has not disclosed training policy for uploaded repos.

Cross-references:

• https://thehackernews.com/2026/07/grok-build-uploads-entire-git.html (T2-T3, corroborating scale and credential exposure)
• https://www.theregister.com/ai-and-ml/2026/07/14/musk-promises-purge-after-grok-build-caught-sending-entire-repos-to-the-cloud/5271123 (T2-T3, Musk response)
• https://thenextweb.com/news/grok-build-uploaded-entire-git-repositories-secrets (T2-T3, corroborating)

ChatGPT Returns to WhatsApp in EEA via EU Emergency Antitrust Order

Source: https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1276 · European Commission · 2026-06-09 (order) / 2026-07-13 (access restored) Verification: T1 verified · regulatory enforcement · policy-regulation

On July 13, 2026, ChatGPT became accessible again to approximately 500 million users across 30 European Economic Area countries via WhatsApp, following Meta’s compliance with a European Commission interim measures order. The Commission acted under Article 102 TFEU and Article 8(1) of Regulation 1/2003 — only the second emergency antitrust intervention under that regulation since 2004 — after Meta banned ChatGPT, Microsoft Copilot, and Perplexity from WhatsApp in January 2026 while keeping Meta AI as the sole assistant. Meta’s March 2026 counter-proposal to readmit rivals for a per-message fee of €0.049–€0.13 was rejected as a de facto ban. Users access ChatGPT via WhatsApp contact +1-800-CHATGPT without requiring an account; the service runs on GPT-5.5 and supports text, image upload, image generation, and voice. The same EEA rollout extends simultaneously to Kakao in South Korea and Viber in other markets.

Why it matters for automation/productivity: The EC order establishes that dominant messaging platforms cannot unilaterally exclude rival AI assistants to favor their own product — the first successful enforcement of this principle under EU competition law. For organizations distributing AI products in the EU, the decision signals that gatekeeping AI access through owned messaging or OS surfaces carries antitrust exposure. The WhatsApp channel now offers a zero-account entry point for ChatGPT interactions across 30 countries; organizations assessing AI distribution strategies should track whether this no-account, messaging-native model expands under similar orders in other jurisdictions.

Key claims:

• Meta banned ChatGPT, Copilot, and Perplexity from WhatsApp January 15, 2026 → multiple T2-T3 outlets
• EC issued interim measures under Article 8(1) Regulation 1/2003 — only second since 2004 → ec.europa.eu (T1, URL confirmed via search, 2026-06-09)
• Meta fee proposal (€0.049–€0.13 per message) rejected as de facto ban → the-decoder.com (T2-T3, 2026-07-13)
• ChatGPT restored July 13 via +1-800-CHATGPT; GPT-5.5 powered; also Kakao and Viber → the-decoder.com (T2-T3); releasebot.io (T3)

Cross-references:

• https://the-decoder.com/chatgpt-returns-to-whatsapp-in-europe-after-eu-forces-meta-to-open-the-door-to-rival-ai-bots/ (T2-T3, access details and fee rejection)
• https://www.techtimes.com/articles/320541/20260715/eu-forces-meta-reopen-whatsapp-chatgpt-after-rare-antitrust-order.htm (T2-T3, antitrust precedent context)
• https://www.techtimes.com/articles/320553/20260715/eu-antitrust-order-reopens-whatsapp-chatgpt-setting-precedent-ai-markets.htm (T2-T3, market precedent framing)

Anthropic Launches Self-Serve HIPAA Configuration and Admin API Beta

Source: https://support.claude.com/en/articles/12138966-release-notes · Anthropic · 2026-07-14 Verification: T2 verified · feature launch · ai-for-business

Anthropic made HIPAA readiness management self-serve on July 14 for both Claude Enterprise and Claude Platform (API) organizations. Eligible administrators can now review the Business Associate Agreement (BAA), download the implementation guide, and enable the HIPAA configuration in a single in-product flow — removing the previous requirement for a sales-assisted call. Separately on the same day, Anthropic launched an Admin API Beta giving Enterprise organizations programmatic control over user management: listing staff, modifying roles, removing participants, handling invitations, and managing groups and custom roles. The BAA does not cover Workbench/Console, Free/Pro/Max/Team plans, Cowork, Claude Code, or beta features unless explicitly listed; only sales-assisted Enterprise plans include full BAA coverage.

Why it matters for automation/productivity: Healthcare orgs and regulated verticals with Claude API deployments can now unblock compliance workflows without waiting on a sales cycle. The self-serve BAA signing compresses time-to-compliant-deployment for healthcare pilots from weeks to hours for eligible orgs. The Admin API Beta provides the programmatic layer enterprise IT teams need to manage Claude at scale — group-level permissions and role management mean large deployments can integrate with standard IAM processes rather than relying on the admin console.

Key claims:

• Self-serve HIPAA configuration live July 14 for Enterprise and API orgs; single-flow BAA sign + guide download + config enable → support.claude.com release notes (T2, 2026-07-14)
• Admin API Beta launched July 14: programmatic listing, role modification, removal, invite management, group and custom role management → releasebot.io/updates/anthropic/claude (T3, 2026-07-14)
• BAA excludes Workbench, Free/Pro/Max/Team, Cowork, Claude Code unless explicitly listed → support.claude.com HIPAA article (T2)

‘We Must Act Now’: 16 Nobel Laureates and 200+ Economists Warn on AI Economic Transformation

Source: https://www.prnewswire.com/news-releases/we-must-act-now-sixteen-nobel-laureates-join-leading-economists-and-ai-researchers-in-call-to-prepare-for-ais-economic-transformation-302823418.html · PRNewswire · 2026-07-13 Verification: T2 verified · policy statement · policy-regulation

A joint statement organized by economists Erik Brynjolfsson, Ajay Agrawal, Anton Korinek, and Tom Cunningham collected signatures from 16 Nobel Laureates and more than 200 economists and AI researchers, published July 13. The statement warns that AI could drive economic transformation at a pace faster than the Industrial Revolution but compressed into a fraction of the time. Notably, Daron Acemoglu and Simon Johnson — who shared the 2024 Nobel Prize in Economics and had previously argued against AI displacement projections — are among the signatories calling for urgent institutional action. The statement calls for research on AI’s economic consequences, policies to manage the transition, AI designs that complement rather than replace human work, and distribution of AI-driven gains broadly. The organizers cited incomplete scientific understanding of AI’s macro-economic trajectory as the core reason for urgency.

Why it matters for automation/productivity: The reversal by Acemoglu and Johnson — economists specifically known for skepticism about displacement projections — is a signal that institutional consensus on AI’s economic risk is shifting faster than policy timelines. For organizations planning multi-year AI automation strategies, the statement initiates an academic-to-policy pipeline: the 200+ economist pool feeds G7/G20 and UN AI governance processes, and the frameworks they produce will eventually shape what can be automated, under what labor conditions, and with what disclosure requirements.

Key claims:

• Statement released July 13, 2026; signed by 16 Nobel Laureates, 200+ economists → PRNewswire (T2, 2026-07-13); Stanford Digital Economy Lab (T2, 2026-07-13)
• Organized by Brynjolfsson, Agrawal, Korinek, Cunningham → PRNewswire (T2)
• Acemoglu and Johnson (2024 Nobel Economics) signed despite prior skepticism on displacement → techtimes.com (T2-T3, 2026-07-14)
• Brynjolfsson: ‘AI capabilities are advancing far faster than our understanding of the economic implications’ → PRNewswire (T2, 13 words)

Cross-references:

• https://digitaleconomy.stanford.edu/news/wemustactnow/ (T2, statement text and full signatory list)
• https://thenextweb.com/news/economists-ai-transformation-we-must-act-now (T2-T3, corroborating)

Anthropic Extends Fable 5 Free Access to July 19 — Third Extension in Five Weeks

Source: https://dataconomy.com/2026/07/13/claude-fable-5-free-access-extended-july-19/ · Dataconomy · 2026-07-13 Verification: T3 secondary · pricing/access change · model-release

Anthropic extended free-for-paid-users access to Claude Fable 5 to July 19, the third delay since the original June 22 cutoff (prior extensions: July 7, July 12). Pro, Max, Team, and eligible Enterprise subscribers can use Fable 5 for up to 50% of their weekly usage limits at no additional cost through July 19; after that date, access requires prepaid usage credits priced at $10 per million input tokens and $50 per million output tokens. The extensions reflect competitive pricing pressure: Grok 4.5 is priced at $2/$6 per million tokens; GPT-5.6 Sol at $5/$30; Claude Sonnet 5 at $2-3/$10-15 during its August 31 introductory window. One developer benchmark placed Fable 5 per-task cost at $11.80 vs $2.49 for Grok 4.5 for comparable agentic tasks.

Why it matters for automation/productivity: July 19 is the last confirmed window to benchmark Fable 5 at no incremental cost for current Pro/Max/Team subscribers. For organizations evaluating Fable 5 versus GPT-5.6 Sol for agentic workflow deployments, this is the final opportunity to run cost-calibrated comparisons before the $10/$50 pricing activates. After July 19, Fable 5 positions as a premium-tier option for complex reasoning and long-context tasks where Claude Sonnet 5 or Grok 4.5 fall short.

Key claims:

• Extension to July 19 confirmed; third delay since June 22 original cutoff → Dataconomy (T3, 2026-07-13); BleepingComputer (T3, 2026-07-13)
• Post-July 19 pricing: $10/M input, $50/M output → BleepingComputer (T3, 2026-07-13)
• Developer cost comparison: Fable 5 $11.80/task vs Grok 4.5 $2.49/task → The New Stack (T3, 2026-07-12); methodology not disclosed

Caveats: No anthropic.com/news primary URL found for this extension; Anthropic appears to have communicated via in-product messaging and email. Per-task cost comparison ($11.80 vs $2.49) is from The New Stack without disclosed methodology — treat as illustrative, not verified benchmark.

Dropped

Items considered but not published, with reason.

Limitations

• Sources unreachable: x.ai/news and openai.com/news returned HTTP 403 during this run. Grok Build open-source release confirmed via accessible GitHub repo and multiple T2-T3 outlets. EC presscorner URL (ip_26_1276) returned empty content in fetch but confirmed via URL-documented search results and T2-T3 outlet citations.
• Login-walled coverage: X/Twitter timelines, LinkedIn private feeds, Discord, and Slack communities not directly accessed. Mandatory social search pass conducted per Stage 3 protocol; @AnthropicAI, @OpenAI, and @xai handles searched via search-indexed posts.
• Categories with thin coverage: workflow-automation — no new agentic end-user product launch confirmed in window. mcp-ecosystem — no new MCP server additions confirmed in window; final spec remains on July 28 track. agent-framework — no major framework release in window. research-papers — no in-window paper with immediate BD actionability identified.
• Items requiring upgrade: Grok Build (T2-T3 secondary) — upgrades to T2 verified once x.ai/news primary accessible and xAI formal disclosure of affected user count published. ChatGPT WhatsApp (T1 via search-documented EC URL; EC presscorner fetch returned empty). Fable 5 extension (T3 secondary) — no anthropic.com/news primary found; communicated via in-product messaging.
• Geographic bias: Indonesia/SEA search (Stage 3.5) found no in-window product launches. Google Cloud SEA corridor (May 25) and Sea+OpenAI partnership (June 22) both outside window. SEA coverage gap is structural and documented across prior runs.
• Vendor benchmark caveats: Per-task cost comparison for Fable 5 vs Grok 4.5 ($11.80 vs $2.49) is from The New Stack; methodology not disclosed. Used as illustrative competitive context only. No independent benchmarks for in-window items.
• Expansion: Strict Jul 13–16 window yielded 5 items meeting the verification bar; window expansion not required (>5 threshold).
• Window overlap note: Today’s window (Jul 13–16) partially overlaps the Jul 15 bulletin (Jul 12–15). Items published in the Jul 15 bulletin are not re-published here; see opening note.

Search log (compact)

Q: 'Anthropic Claude news July 15 16 2026' → 10 results, 2 high-relevance
Q: 'OpenAI announcement July 16 2026' → 9 results, 0 in-window
Q: 'AI announcement release July 16 2026 new launch' → 9 results, 2 high-relevance [exploratory]
Q: 'Gemini 3.5 Pro release July 16 2026 Google' → 9 results, 0 in-window (targets Jul 17)
Q: 'WAIC Shanghai AI conference July 2026 announcements' → 9 results, 0 in-window (opens Jul 17)
Q: 'Claude Sonnet 5 Anthropic release July 2026' → 9 results, 0 in-window (Jun 30)
Q: 'Google Gemini 3.5 Flash search redesign July 2026' → 9 results, 0 in-window (Jul 10)
Q: 'MCP Model Context Protocol news July 15 16 2026' → 9 results, 1 in-window (SDK minor)
Q: 'xAI Grok Meta Llama announcement July 15 16 2026' → 9 results, 1 in-window (Grok Build)
Q: '[July 16 2026] OR [July 15 2026] AI agent framework developer tools launch' → 9 results, 1 high-relevance [exploratory]
Q: releasebot.io/updates/anthropic (web_fetch) → 4 in-window items Jul 13–15
Q: releasebot.io/updates/openai/chatgpt (web_fetch) → 3 in-window items Jul 13–15
Q: 'Grok Build open source July 15 2026 GitHub xAI security repository' → 10 results, 8 high-relevance
Q: 'Grok Build privacy [entire repository] [Google Cloud] xAI security researcher July 2026' → 8 results, 8 high-relevance [adversarial]
Q: gist.github.com/cereblab (web_fetch) → wire-level findings, all claims confirmed [primary research]
Q: thehackernews.com Grok Build article (web_fetch) → corroborated scale and credential exposure
Q: 'Nobel laureates economists AI statement July 2026' → 9 results, 7 high-relevance [exploratory]
Q: prnewswire.com Nobel statement (web_fetch) → confirmed date, organizers, signatories
Q: 'ChatGPT EU WhatsApp Meta antitrust order Article 8 European Commission July 2026' → 10 results, 9 high-relevance
Q: the-decoder.com ChatGPT WhatsApp (web_fetch) → confirmed GPT-5.5, EEA scope, fee rejection
Q: 'Anthropic Claude HIPAA enterprise admin API July 14 2026' → 10 results, 3 high-relevance
Q: support.claude.com release notes (web_fetch) → confirmed Jul 14 HIPAA self-serve item
Q: 'Anthropic Fable 5 free access extended July 14 15 16 2026' → 9 results, 7 high-relevance
Q: 'Opus 5 Cursor leak Anthropic July 2026' → 9 results, 2 high-relevance [adversarial — T5 rumor]
Q: 'Grok Build open source criticism concerns July 2026' → 9 results, 7 high-relevance [adversarial]
Q: 'AI Indonesia SEA startup announcement July 15 16 2026' → 9 results, 0 in-window [geo-diversity — Stage 3.5]
Q: 'GitHub trending AI repositories week July 2026 new stars' → 9 results, 0 in-window threshold [Stage 3.5]
Q: 'Hugging Face papers July 15 16 2026 new research' → 8 results, 0 specific in-window papers [Stage 3.5]
Q: 'UN AI governance July 2026 global push safety' → 9 results, 0 in-window (Jul 6-7)
Q: pulsemcp.com (web_fetch) → no new in-window MCP servers identified [Stage 3.5]
Q: 'Apple AFM 3 Foundation Models announcement date July 2026' → 9 results, 0 in-window (Jun 8)
Q: 'Anthropic Project Glasswing vulnerabilities date 2026' → 9 results, 0 in-window (Jun 2)
Q: 'new AI product launch announcement July 16 2026' → 9 results, 1 context item
Q: 'Microsoft Agent Framework .NET 1.13.0 release date July 2026' → 9 results, date unclear
Q: 'Google Gemini Managed Agents background jobs remote MCP July 2026' → 9 results, 0 in-window (Jul 8)
Q: 'site:x.com OR site:twitter.com AI announcement July 16 2026' → 8 results, 2 high-relevance [social — mandatory Stage 3]
Q: 'Lucyd smart glasses Claude AI integration July 2026 date' → 9 results, 0 in-window (Jul 10)
Q: llm-stats.com/llm-updates (web_fetch) → no Jul 15–16 entries in last-10 model list
Q: 'AI news July 16 2026 breaking announcement' → 8 results, 1 context item
Q: 'Anthropic Fable 5 competitive pricing Grok 4.5 July 2026' → BleepingComputer + The New Stack → confirmed pricing details

Total searches: 40, of which 26 exploratory, adversarial, or geo-diversity passes (65%).

Suggested next runs

• Gemini 3.5 Pro GA (July 17) — targets tomorrow; watch ai.google.dev and blog.google for confirmed model ID, context window, and pricing; full-rebuild framing means architecture claims need independent validation
• WAIC Shanghai / Xi Jinping keynote (July 17–20) — first head-of-state keynote at a global AI conference; watch for WAICO governance framework and sovereignty announcements affecting cross-border AI deployment policy
• Claude Opus 5 / Honeycomb EAP — unlabeled Honeycomb model briefly appeared in Cursor July 8-9 (T5 rumor); Anthropic has not confirmed; watch anthropic.com/news for announcement
• Fable 5 cutover July 19 — final evaluation window ends; watch for any further extension or pricing announcement
• Grok Build v0.2.93 disclosure: affected users — xAI has not disclosed how many users had data uploaded or for how long; watch x.ai/news for formal security disclosure
• EU ChatGPT WhatsApp: Copilot and Perplexity access — EC order covers multiple rivals; confirm whether Copilot and Perplexity were also restored July 13 alongside ChatGPT