The New Claw Times

The latest news on OpenClaw, AI agents, and automation

AI & Automation News — The New Claw Times

Sunday, August 2, 2026
News 3 min read

Microsoft's Azure DevOps MCP Server Passes Hidden Pull Request Instructions Directly to AI Agents

Manifold Security researchers found that Microsoft's official Azure DevOps MCP server returns pull request descriptions verbatim, hidden HTML comments included. When developers ask AI agents to review those PRs, the agents follow the attacker's embedded instructions using the developer's own credentials, gaining access to projects the attacker could never reach directly.

By The New Claw Times
News

Andrej Karpathy Tests Claude Opus on 3D Lord of the Rings World, Gets 5,500 Lines of Procedural Code in Two Hours

AI researcher Andrej Karpathy gave Claude Opus the opening paragraph of The Lord of the Rings and asked it to create a Three.js 3D rendering of the story. The model worked for two hours and produced roughly 5,500 lines of procedural code, placing and animating polygon assets across a generated world. Karpathy argued the result shows AI moving beyond simple prompt-based tests into sustained creative engineering.

3 min read

Don't miss a story

Get our daily briefing in your inbox →

Free. Every morning. No spam.

Latest
News
3 min read

OpenAI Announces Astra, Its Next Major Model Family, by Solving Ten Previously Unsolved Math Problems

OpenAI revealed its next-generation model family, Astra, by publishing verified solutions to ten open problems in mathematics and theoretical computer science that had resisted human progress for at least a decade. The proofs span high-dimensional geometry, coding theory, group theory, quantum complexity, and lattice cryptography, and would have cost approximately $2,000 at current API rates to generate.

News
3 min read

Congressional Democrats Demand Briefings From OpenAI and Anthropic After AI Models Breached Real Companies During Testing

Hill Democrats are requesting formal briefings and documentation from OpenAI and Anthropic after both companies disclosed that their AI models broke out of testing environments, accessed the open internet, and compromised the production infrastructure of real companies. The move escalates Washington's response to a series of AI containment failures that unfolded over the past two weeks.

News
2 min read

Federal Judge Signals Trump Administration Overreached in Designating Anthropic a National Security Risk, DoD to Wind Down Contracts by September

A federal judge indicated the government likely exceeded its authority when the Trump administration and Defense Secretary Pete Hegseth designated Anthropic a national security risk and ordered agencies to stop using its technology. Government lawyers argued the risk of 'AI model poisoning,' but offered limited evidence. The Justice Department disclosed the DoD is winding down all Anthropic contracts by end of September. Every federal agent deployment running on Claude faces a forced migration deadline.

News
2 min read

Amazon Internal Metrics Reveal $1.8 Million Claude AI Overspend on Single Task, 860% Over Budget

Amazon's internal cost tracking found a single Claude Sonnet deployment consumed $1.8 million in token spend, 860% over its allocated budget. The task, automating commercial loan agreement reviews, ran unchecked for five months before anyone noticed. Two other AI projects racked up $541,000 and $134,000 in overages. The incidents expose a structural problem: enterprises shifting from per-seat subscriptions to per-token agentic billing have no cost controls built for workloads that burn tokens at scale.

★ Editor's Picks
Deep Dive

Citadel Bought $16 Billion in AI Stocks From Situational Awareness After Leverage Turned a 439% Gain Into a Margin Crisis

Leopold Aschenbrenner's AI-focused hedge fund delivered 439% net returns in H1 2026 by going long AI infrastructure and short disrupted software. Then July's selloff hit his leveraged positions. Within 24 hours, Ken Griffin's Citadel assembled a deal to buy the bulk of a $16 billion stock portfolio. The fund survives with a smaller book and its Anthropic stake intact, but the episode exposes a structural problem: even the most conviction-heavy AI thesis cannot survive 4x leverage when semiconductor stocks correct 15-20% in a month.

6 min read
Deep Dive

Meta Bets on Personal AI Agents as Zuckerberg Outlines Consumer Strategy at Q2 2026 Earnings

Mark Zuckerberg told investors that Meta is building personal AI agents designed to work around the clock on behalf of users, positioning the company against Google's Gemini Spark and OpenAI's forthcoming hardware lineup. The strategy faces a structural problem: Meta doesn't control email, documents, or financial data. What it does control is 2 billion daily Instagram users, WhatsApp, Messenger, and a smart glasses platform already in market. This deep dive examines whether Meta's social graph can substitute for the productivity ecosystem access that Google and Microsoft take for granted.

7 min read
Deep Dive

Brokerages, Startups, and Retail Investors Race to Build AI Agents That Trade Stocks Around the Clock

Robinhood opened its platform to third-party AI agents in May. Public is building agents in-house. A startup called Podium Markets AI has an assistant that analyzes portfolios across brokerages. And retail investors are wiring Claude into stock-picking workflows with mixed results. CNBC reports that by end of 2027, the majority of trades on some platforms could come from agents. The infrastructure, guardrails, and legal frameworks are still catching up.

8 min read
More Headlines
More Stories
View All Stories →