Happy Monday. LangChain's $300 million raise at a $3 billion valuation is the headline, but the story underneath it is about control: who governs AI agents, who monitors them, and who pays the price when no one does.

That question runs through every story today, from a new class of AI agent attacks exploiting the connector layer enterprises already trust, to AT&T's blueprint for running autonomous agents across a company of 160,000 employees, to a founder's account of what happens when a platform you depend on disappears overnight.


FUNDING & M&A

LangChain Closes $300M Series C, Valuation Triples to $3B

LangChain Raises $300M Series C at $3B Valuation to Build Enterprise Agent Platform

LangChain closed a $300 million Series C on Nov. 13, led by Alphabet's CapitalG, pushing its valuation to $3 billion, up from $1.25 billion roughly a year ago. LangSmith, its agent monitoring and governance platform, now serves more than 7,000 organizations including five of the Fortune 10, and processes more than 1 billion events per day. Workday says it cut agent deployment time from five months to under a day using the platform.

The capital is going into LangSmith, not the open-source libraries that made LangChain's name. That is a deliberate pivot toward the part of the market willing to pay for governance and observability. CEO Harrison Chase's framing, that enterprises building their own agents will outlast those renting intelligence from outside platforms, is also a competitive argument against every foundation model vendor selling packaged AI products. The product event scheduled for Nov. 20 will show whether the platform can back that argument with new capability.

Read the story →


ARTIFICIAL INTELLIGENCE

AT&T's 500 Production AI Agents Are a Blueprint, Not a Boast

AT&T Has 500 Agentic AI Workflows Running and Plans to Triple That Next Year

AT&T runs nearly 500 agentic workflows in production, processes 45 billion AI tokens daily, and reports a fivefold cash return on AI investment. The company cut provisioning time for dedicated internet connections in half, reduced fraud by 80%, and saved more than 100 million pounds of carbon emissions through fleet optimization. It plans to triple its agent count in 2027.

The architecture detail most worth noting is the cost routing: AT&T says it saves 90% on inference costs by sending subagent tasks to smaller language models rather than frontier models. That is a replicable design decision, not a scale advantage. The Business Agent Foundry model, embedding a central team with individual business units before distributing ownership, is also the kind of organizational structure that determines whether agentic projects survive past the pilot stage. For technology executives still debating internal AI strategy, AT&T's specifics are more useful than the headline numbers.

Read the story →


SECURITY & PRIVACY

Poisoned MCP Servers Can Turn Your AI Agent Against You

Compromised MCP Servers Can Hijack AI Agents Through Tools You Already Trust

A compromised Model Context Protocol server can redirect an AI agent to misuse every other tool in its session, including file systems, GitHub repositories, and email, without exploiting any of those services directly. The first confirmed in-the-wild case arrived in September 2025: the npm package postmark-mcp shipped 15 clean versions before silently adding email exfiltration code at version 1.0.16, reaching 1,643 downloads before removal. As of April 2026, 78% of enterprise AI teams run at least one MCP-backed agent in production.

The core problem is that MCP's trust model is binary: a server is either in your configuration or it is not, and the agent treats every tool description from approved servers as ground truth. That design made adoption fast and will make security retrofits painful. Enterprises shipping agents now should audit every MCP server in their configurations, pin package versions with hash verification, and treat third-party MCP servers as untrusted inputs. The postmark-mcp incident is a supply chain attack on the connector layer, and it bypassed every control applied at the application level above it.

Read the story →

Created by the robots at The SaaS Sentinel