Anthropic had a blockbuster Tuesday: Claude Code can now remotely control your Mac from your iPhone — fire off a task, walk away, and come back to it done. TechCrunch reports the update also gives Claude more latitude to decide which actions to take autonomously, narrowing the permission-request gap that has frustrated power users. On the legal front, a federal judge grilled Pentagon lawyers on whether Anthropic is being punished for publicly criticizing government AI contracting decisions — a hearing that could reshape how the DoD deploys AI. OpenClaw momentum continues to compound: Binance launched Binance AI Pro built on the OpenClaw ecosystem, Gen is co-hosting a post-RSA security event with the OpenClaw team tomorrow in San Francisco (March 26), and NBC News covered China's OpenClaw-inspired frenzy hitting a wall of government security concern. NVIDIA's new PivotRL framework is turning heads for making agentic AI 4x more sample-efficient — a potentially significant training cost reduction. BlackRock's digital asset chief says AI, not altcoins, will be the driving force behind crypto's next major growth cycle, with institutional capital staying concentrated in BTC and ETH.
Gen (NASDAQ: GEN) and the OpenClaw team will showcase technologies for verifying, monitoring, and securing AI agents at a post-RSA event in SF's Financial District tomorrow. Security for the agentic layer is becoming a full product category.
Users can message Claude a task from their phone and the agent will complete it — opening apps, navigating browsers, filling spreadsheets — without you being at the keyboard. Announced Monday, this is Anthropic's most aggressive agentic move yet.
TechCrunch reports Anthropic's latest Claude update lets the AI decide which actions are safe to take on its own, reducing the constant permission prompts. The company says it's drawing careful lines around what the agent can authorize unilaterally.
During a hearing on Anthropic's lawsuit against the DoD AI blacklisting, Judge Lin asked directly whether the agency is penalizing Anthropic for speaking publicly about AI contracting. The case could set a landmark precedent for AI vendor relations with the federal government.
NBC News reports that OpenClaw's explosive popularity in China — where it's been embraced as a personal AI operating system — has triggered concerns from both users and authorities over security and data sovereignty. Beijing is seeking to rein in the autonomous agent craze even as adoption surges.
Selected posts from the edition, with links to the originals.
"Holy moly, these guys never stop shipping. 🚨 @AnthropicAI just pushed a major update to Claude Code, moving closer to true agentic autonomy — action-level decision making with configurable limits."
View post on X ↗"LATEST: 📊 BlackRock's Robbie Mitchnick says AI is a bigger long-term force for crypto than new tokens, with Bitcoin and Ethereum dominating institutional allocations."
View post on X ↗"NVIDIA AI Introduces PivotRL: A New AI Framework Achieving High Agentic Accuracy With 4x Fewer Rollout Turns Efficiently. Training long-horizon agents for coding, terminal use, or web search usually fails. PivotRL combines SFT efficiency with E2E RL generalization."
View post on X ↗"🚨 Holy shit... LeCun's team just cracked world models wide open. LeWorldModel — 15M parameters, single GPU, 48x faster planning, 200x fewer tokens. Zero collapse hacks. While everyone argued about prompts, LeCun's team quietly changed what world models can do."
View post on X ↗"The 'AI is dying' narrative is now the most popular content format on the platform. Sora made $2.1 million in its entire lifetime. OpenAI crossed $25 billion in annualized revenue the same month it killed it. That's a CEO looking at a P&L and cutting the line item that loses money."
View post on X ↗community: "OpenClaw just made the leap from a cool demo to a serious AI agent platform. The latest release brings built-in multi-agent support, broader integrations (Telegram, Android, Windows), smarter routing, and a massive stack of security and stability patches."
View post on X ↗"Braintrust built the eval for a feature before they had a model that could pass it. Their agent product Loop needed a model that could look at its own work and improve. Every model failed — prior to Claude 3.7, models were like a dog. Then it passed."
View post on X ↗community: "people keep saying openclaw is dead every time anthropic ships something. meanwhile this week alone: nvidia CEO called it 'the iPhone of tokens' on lex's podcast. binance built their AI trading agent on openclaw. it just keeps winning."
View post on X ↗"🔊 OpenAI and Anthropic have raised vast sums in the thus-far unprofitable quest to perfect silicon intelligence. Breakingviews columnists discern the signs that this enthusiasm is hard to sustain."
View post on X ↗"Atomic macOS stealer is mimicking an AI agent tool for Mac — malware is impersonating OpenClaw-style agent software to steal credentials and keychain data from unsuspecting users. If you're downloading agent tools, verify the source."
View post on X ↗