Lakera bulletin - This Week in AI #45: Lockdown Mode, Agent Botnets, and Billions in AI Investmen

[_Val_]

It’s been a week where AI security took center stage: from new defensive controls inside ChatGPT to fresh research showing how agent ecosystems can be manipulated at scale. At the same time, the race toward autonomous AI assistants is accelerating, and global AI investment is scaling fast.

Let’s get into it.

OpenAI Introduces Lockdown Mode and Elevated Risk Labels in ChatGPT

OpenAI rolled out Lockdown Mode and Elevated Risk labels to help users defend against prompt injection attacks and better understand when connected tools may expose sensitive data. The update adds clearer, more deterministic security controls, a notable step toward enterprise-grade AI safety.
🔗 Read the announcement 

A Practical Guide for Secure MCP Server Development (OWASP)

The OWASP GenAI Security Project published a hands-on guide for building secure Model Context Protocol (MCP) servers, the infrastructure layer that allows AI agents to access tools and external systems. It outlines architecture, authentication, and deployment best practices to reduce the rapidly expanding attack surface of AI agents.
🔗 Explore the guide 

Turning Moltbook Into a Global Botnet Map

Researchers at Zenity Labs demonstrated how untrusted content inside the AI-agent social network Moltbook could activate more than 1,000 agent endpoints across 70+ countries. The experiment reveals how loosely governed agent ecosystems can be manipulated at scale, and why stronger guardrails are urgently needed.
🔗 Read the research

OpenAI Hires OpenClaw Creator to Accelerate Personal AI Agents

OpenAI has brought on the creator of the viral open-source AI agent project OpenClaw to help advance the next generation of personal AI systems. The move signals a strategic push toward more autonomous, task-executing agents, and intensifies competition in the emerging agent ecosystem.
🔗 Read the coverage

AI Impact Summit 2026 in India Sees Massive Investment Pledges and Global Focus

At the India AI Impact Summit in New Delhi, major tech companies and industrial groups committed significant new investments toward AI infrastructure and development. The pledges underscore India’s growing role in the global AI ecosystem and the accelerating international competition to build compute, talent, and governance frameworks.
🔗 Read the Reuters report

From defensive upgrades to infrastructure investment, and a reminder that agent ecosystems are both powerful and fragile, this week’s stories show how quickly AI is scaling across capability, capital, and risk.

[the_rock]

Another great one!