Scale AI support on AWS, see how July 9

Customer expectations keep rising. Support budgets don't. On July 9, Fin and AWS are hosting a live executive session on how leading enterprises close that gap: scaling AI-powered support while simplifying how they buy it.

You'll see how to resolve an average 76% of conversations with Fin on AWS enterprise-grade infrastructure, procure through AWS Marketplace to put committed cloud spend to work, and turn the Fin and AWS collaboration into lower support costs. Register for the live session to see how.

Save your spot

Today:

• Anthropic Restores Fable 5

• Gemini Spark Lands on macOS

• Microsoft Opens Frontier Company

• Z.ai Launches ZCode

• Claude Code Adds Artifacts

Claude Fable 5 Returns After Temporary U.S. Export Ban

After an 18-day disruption involving direct federal intervention, Anthropic has officially reinstated its flagship AI models, Claude Fable 5 and Mythos 5. The models were temporarily pulled offline globally following emergency U.S. export controls triggered by concerns over their potential cybersecurity capabilities.

Important Details:

• The Cybersecurity Alarm: The ban was initiated after researchers discovered a "jailbreak" method that bypassed Fable 5's safeguards, which raised fears that the model could be used to uncover and exploit critical software vulnerabilities.

• Enhanced Guardrails: Anthropic worked alongside government officials and partners to implement an improved safety classifier. The models can now better detect and block high-risk cybersecurity prompts while safely allowing legitimate programming and research requests.

• Availability Restored: Fable 5 is now back online across the Claude Platform, Claude.ai, and Claude Code. Pro, Max, and select Enterprise users are receiving promotional access to Fable 5 for up to 50% of their weekly limits through July 7, 2026.

• Deeper Government Collaboration: Moving forward, Anthropic agreed to provide the U.S. government with pre-release access to evaluate the safety of future frontier models and committed to rapidly sharing new threat intelligence.

Gemini Spark Gets a Massive Upgrade: macOS App and Real-Time Tracking

Google is rolling out a sweeping set of June 2026 updates for Gemini Spark, effectively transforming it from a standard chat interface into an automated, multi-agent desktop assistant capable of complex actions and live monitoring.

Important Details:

• macOS Desktop Experience: Gemini Spark is now available as a macOS app (currently in Beta for Google AI Ultra subscribers). It can tackle heavy lifting directly on your local machine, such as bulk-sorting downloads into specific folders or generating budget spreadsheets from local invoice files.

• Remote Task Execution: Users will soon be able to assign multi-step tasks to their Mac via their phone while away from the computer—like instructing Spark to find a specific local sales report, extract the revenue numbers, and email them.

• App Integrations & Custom Connectors: Spark now natively connects to Google Keep and Tasks, allowing it to turn scattered notes into actionable checklists. It also integrates with Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals, and it supports custom Model Context Protocol (MCP) so developers can plug in their own proprietary tools.

• Real-Time Tracking: You can now instruct Spark to autonomously monitor live events—such as sports scores, stock price thresholds, or breaking news—and send proactive alerts without you ever needing to hit refresh.

Microsoft Launches 'Frontier Company' to Drive Enterprise AI

Microsoft has announced the formation of Microsoft Frontier Company, a new operating business backed by a massive $2.5 billion investment. Led by newly appointed President Rodrigo Kede Lima, the division is designed to help enterprise customers turn AI experimentation into measurable business outcomes through what Microsoft calls "Frontier Transformation."

Important Details:

• Massive Engineering Deployment: The initiative is embedding 6,000 Microsoft industry and engineering experts directly within customer organizations to co-design, deploy, and continuously fine-tune agentic AI workflows.

• Intelligence + Trust: Microsoft is heavily emphasizing data security, guaranteeing that a customer's proprietary data and intellectual property will never be used to train external models, thereby protecting their unique competitive advantage.

• Model-Agnostic Flexibility: Rather than forcing vendor lock-in, the Frontier Company utilizes an open, heterogeneous AI platform. Customers can choose to route workflows through whichever model fits the task best—whether that's from OpenAI, Anthropic, Microsoft AI, or an open-source alternative.

• Early Success: The hands-on engineering approach is already yielding fast, measurable results for early flagship partners like the London Stock Exchange Group (LSEG), Unilever, Novo Nordisk, and Land O’Lakes.

🧠RESEARCH

Orca: The World is in Your Mind

Orca is an AI model trained to predict what happens next across video, language and actions. It learned from 125,000 hours of video and 160 million described events, building one representation of how the world changes. Tests found this foundation supported text, image and robot tasks better than comparable models.

Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs

Researchers trained language models to judge how well they performed and express uncertainty more honestly. Their method rewards answers and accurate self-assessment, then rewrites confidence scores into clear wording. Across ten tasks, it preserved accuracy, improved uncertainty reporting and beat reinforcement learning—training through rewards—by up to 63 percent.

ASPIRE: Agentic /Skills Discovery for Robotics

Nvidia created ASPIRE, a system that writes robot programs, studies failures and saves successful fixes as reusable skills. It improved results by up to 77 percentage points and completed unseen, multi-step tasks far more often than earlier methods. Skills learned in computer simulations also reduced programming work on physical robots.

📲SOCIAL MEDIA

🗞️MORE NEWS

Z.ai Launches ZCode for GLM-5.2 Z.ai released ZCode, its official coding workspace for GLM-5.2, on macOS, Windows and Linux. It supports existing subscriptions and API keys, while GLM Coding Plan subscribers receive 1.5 times their normal usage allowance.

Claude Code Can Turn Work Into Interactive Pages Claude Code can now publish its results as private, interactive web pages that update as work continues. These “artifacts” can display code changes, dashboards, comparisons and progress reports, but they cannot store information or connect to online services while being viewed.

How Devin Protects Customer Data Cognition says Devin encrypts stored and transmitted information, limits access to its production systems and has passed an independent SOC 2 Type II security review. Customer data is not used to train its AI unless the customer chooses to allow it, and Enterprise data is never used for training; however, users should still review Devin’s code because it can introduce bugs or security problems.

Anthropic Explores a Custom AI Chip With Samsung Anthropic has reportedly begun early discussions with Samsung about manufacturing its own AI server chip, but its design, power and purpose have not been decided. A custom chip could give Anthropic more control over costs and supply while reducing its dependence on Nvidia, though the company says chips from Nvidia, Google and Amazon will remain important.

OpenAI Considers Giving the U.S. Government a 5% Stake OpenAI is reportedly discussing giving the U.S. government a 5% ownership stake, possibly through a public investment fund that would share AI-generated wealth with citizens. The proposal remains an early idea, could require approval from Congress and may depend on other major AI companies offering similar stakes.