The New Claw Times

The latest news on OpenClaw, AI agents, and automation

AI & Automation News — The New Claw Times

Saturday, July 18, 2026
News 3 min read

Brex Open-Sources CrabTrap, an LLM-as-a-Judge Proxy That Learns Agent Security Policy from Real Network Traffic

Brex released CrabTrap, an open-source proxy that sits between AI agents and every outbound network request. Instead of writing security policies in advance, CrabTrap observes agent behavior, uses an LLM judge on the 3% of requests that fall outside known patterns, and auto-generates policy from real traffic. CEO Pedro Franceschi says existing guardrails forced a choice between capability and safety. CrabTrap's transport-layer approach is framework-agnostic and works with any agent harness, including OpenClaw.

By The New Claw Times
News

Composio Adds Ollama and Hugging Face Integrations for Hermes and Claude Cowork Agent Frameworks

Agent integration platform Composio shipped two new MCP-based toolkits overnight, connecting the Hermes autonomous agent framework to Ollama's local model runtime and Anthropic's Claude Cowork to Hugging Face model repositories. The releases extend Composio's coverage to 20,000+ tools across more than 1,000 applications, reinforcing its position as middleware for the fragmented agent framework market.

2 min read
News

PentestCode Brings Autonomous AI Agents to Offensive Security with 18 Integrated Penetration Testing Tools

PentestCode is an open-source AI agent purpose-built for penetration testing. A hard fork of OpenCode, it coordinates 13 specialized agents across reconnaissance, exploitation, and post-exploitation workflows using 18 integrated security tools. Token costs run $5 to $50 per engagement. The tool automates the tactical decision tree of methodical enumeration, not creative exploit development.

3 min read
News

Anthropic in Early Talks to Lease Compute from Meta in Deal Worth Up to $10 Billion

Anthropic is in very preliminary talks to lease GPU compute from Meta's data centers, according to CNBC and The New York Times. The potential deal could be worth about $10 billion. It follows Anthropic's recent agreement to use SpaceX's Colossus 1 data center and signals that frontier AI labs are increasingly dependent on hyperscaler compute partnerships rather than building their own infrastructure.

2 min read

Don't miss a story

Get our daily briefing in your inbox →

Free. Every morning. No spam.

Latest
News
3 min read

a16z Leads $20M Seed for Runta, Betting Agent Infrastructure Needs Its Own Operating System

Andreessen Horowitz is leading a $20 million seed round for Runta, a startup founded by former Cloudflare and Kong engineer Guanlan Dai. The company is rebuilding the execution layer where AI agents run, arguing that agents need a full operating system with security controls built in, not another sandbox cloud. a16z partner Martin Casado says the shift from hosting software to hosting agents is the largest transition in computing since virtualization.

★ Editor's Picks
Deep Dive

a16z, Accel, Greylock, and NEA Bet Over $110M on AI Agent Governance in a Single Week

Three agent governance startups raised over $110 million in disclosed funding within days of each other in mid-July 2026. Oak took $60 million to unify identity governance for humans, machines, and AI agents. Valarian closed $50 million for sovereign AI control planes. And a16z backed Ode, a startup that treats agent management like parenting. The convergence reveals a market thesis taking shape in real time: the bottleneck to enterprise agent deployment is not capability. It is control.

7 min read
Deep Dive

Voice AI Startups Raise $124M as Enterprise Phone Systems Become the Next Agent Frontier

While coding agents and chat interfaces dominate AI headlines, voice AI startups have quietly raised $124 million in fresh capital. Rime closed a $24M Series A led by M13 with Twilio Ventures participating, and Paris-based Gradium pulled in $100M with Nvidia backing. The capital is chasing a market that chat widgets cannot reach: the billions of enterprise phone calls that still run through legacy systems in healthcare, banking, and hospitality.

7 min read
Deep Dive

The Agent Production Gap: Microsoft, Fujitsu, and Oracle Are Building the Tooling Nobody Had Six Months Ago

Microsoft released a seven-lesson open-source course on July 14 that walks developers through every stage of building production AI agents, from design through evaluation, deployment, data sovereignty, tool governance, and cross-framework interoperability via the A2A protocol. The course arrives alongside Fujitsu's auto-generating multi-agent framework, Oracle's pro-code Agent Studio tooling, and three independent evaluation framework releases, all within a single week. The pattern is clear: the industry has moved past 'can we build agents?' and is now solving 'how do we ship them safely?'

8 min read
More Headlines
More Stories
View All Stories →