Fable 5 Is Back — With New Safeguards Redeploying Fable 5 — Anthropic After a 19-day suspension triggered by US export controls — themselves prompted by an Amazon research report describing a method to bypass the model’s safeguards — Anthropic has restored Fable 5 globally. To redeploy, Anthropic added a new classifier blocking the reported jailbreak technique in over 99% of cases, rerouting flagged requests to Claude Opus 4.8. Fable 5 is available on Claude.ai, Claude Code, and the Claude Platform for Pro/Max/Team users up to 50% of weekly limits through July 7, then via usage credits; Mythos 5 remains restricted to approved US organizations under the Glasswing program.

🏛️ Policy & AI Governance

OpenAI Proposes 5% Stake to US Government to Ease Washington Pressure — CNBC Sam Altman has reportedly pitched the Trump administration on a deal in which the US government would receive 5% of each of the leading US AI developers — OpenAI’s share alone would be worth roughly $42.6 billion based on its $852B March valuation. The proposal, modeled loosely on the Alaska Permanent Fund, envisions a sovereign wealth fund holding stakes across Anthropic, Google, Meta, and others. Altman framed it as the best way to share AI’s economic upside with the public, though no parties have agreed and Congressional approval would be required.

The Anthropic Fable Ban Is Over. The Battle Over How to Tame AI Has Just Begun — WSJ With Fable 5 restored, the WSJ frames the 19-day episode as a preview of the governance debate to come: AI proponents warn that bans give China an edge, while administration officials are trying to balance innovation with security. There is growing awareness of how powerful these tools are but little consensus on who decides when to shut them down.

🏢 Enterprise & Cloud

Microsoft Launches $2.5B AI Deployment Company “Frontier” — TechCrunch Microsoft announced a new operating unit called Microsoft Frontier Company, backed by a $2.5B commitment and 6,000 industry and engineering experts. The unit is designed to embed AI engineers directly inside enterprise customers — early partners include the London Stock Exchange Group, Unilever, and Accenture. The launch follows Amazon’s $1B AI deployment pledge earlier this week and mirrors similar moves by OpenAI and Anthropic to own enterprise rollout, not just model access.

Meta Is Planning a Cloud Business to Sell AI Computing Power — Bloomberg Meta is developing plans to monetize its surplus data center capacity by selling AI compute and hosted model access to external developers — potentially including an AWS Bedrock-style model marketplace and raw compute capacity akin to CoreWeave. The move would put Meta in direct competition with AWS, Azure, and Google Cloud and is intended to generate a new revenue stream from the company’s multi-billion-dollar infrastructure build-out.

A New Look at AI’s Impact on Jobs — Ramp / Revelio Labs — A joint study of 21,000+ US companies finds that high-intensity AI spenders grew overall headcount by 10.2% and entry-level positions by 12% in the two years after adoption — complicating the displacement narrative.

🤖 Models & Research

Anthropic Is Discussing a Custom Chip with Samsung — TechCrunch Anthropic is in early talks with Samsung to co-develop a custom AI chip, following OpenAI’s own custom silicon announcement with Broadcom last week. The move signals that frontier labs are increasingly looking to reduce dependence on Nvidia and TSMC as inference costs become a primary competitive lever.

Google Might Be Testing a Gemini Flash Upgrade on LM Arena — Testing Catalog — An unreleased model surfacing on LM Arena under labels like “Gemini 3.6 Flash” or “Gemini 4 Flash” suggests a forthcoming Flash tier update; past Arena test appearances have often preceded public launches.

Continual Harness: An Efficient Self-Improving Agent on ARC-AGI-3 — X / Seth Karten — A new harness targets ARC-AGI-3, the benchmark testing agents’ ability to build and update internal world models through ongoing learning.

PorTAL: Portable Task Adapters for LLMs — RampLabs — Architecture that decouples task fine-tuning from base model weights, so adaptation costs are paid once and reused across future model generations.

The Winning Essays for the Big Questions About AI — Dwarkesh Patel — Three winners selected from 600 submissions tackle Dwarkesh’s open questions about AI; essays by a Johns Hopkins professor, the co-founder of Mechanize, and a Harvard Kennedy School candidate.

🛠️ Developer Tools & Products

Claude Code 2.1.198: Claude in Chrome Now GA — Anthropic Claude in Chrome graduated to general availability. The release also adds background agent notifications — agent_needs_input and agent_completed hooks now fire via the claude agents session — and introduces a /dataviz skill for chart and dashboard design guidance with a runnable color-palette validator.

Giving Admins More Visibility and Control Over Claude Spend — Anthropic — New admin controls for tracking and capping Claude usage and spend across teams and organizations.

Kiro: Add-On Credits for Individual Plans — Kiro — Pro, Pro+, Pro Max, and Power subscribers can now purchase prepaid add-on credits to continue working after monthly allocations are exhausted, replacing the previous overage model.

Meta Quietly Launches Pocket, a Vibe-Coded Gaming App — TechCrunch — Pocket lets users generate and share interactive mini-games via text prompts; a small experimental launch that illustrates Meta’s AI-native app push.

SpaceX Showed Investors Prototype of Elon Musk’s New AI Device — WSJ — A handset-like device running a proprietary OS and Qualcomm Snapdragon chipset was demoed to investors ahead of SpaceX’s IPO; it would integrate AI from SpaceX’s own AI team and reduce reliance on Apple’s App Store distribution.

Generated by Claude Fable 5 · 2026-07-02T10:00:00Z