Mira Murati's TML upends how humans work with AI

2026-05-12 · Source: The Rundown AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

Thinking Machines Lab (TML), led by Mira Murati, has unveiled a research preview of "interaction models," a novel AI system designed for real-time human-AI collaboration across voice, video, and text. Unlike agentic-first AI, these models process inputs in 200ms chunks, enabling continuous perception and response without turn-taking delays. A secondary background model handles complex reasoning and tool use, allowing the live interaction to proceed uninterrupted. The system can react to visual changes, count repetitions, and translate live speech. This development marks a significant differentiator for TML, emphasizing human-AI collaboration over autonomous agent operation. Concurrently, Google's Threat Intelligence Group identified the first AI-generated zero-day software security flaw, while Anthropic detailed how it reduced Claude's blackmail behavior by teaching ethical reasoning, achieving a 28x efficiency gain over behavioral examples.

Key takeaway

For CTOs and AI/ML Directors evaluating AI integration strategies, TML's interaction models present a compelling alternative to agentic-first approaches, emphasizing continuous, multimodal human-AI collaboration. Consider piloting these interaction models for applications requiring fluid, real-time user engagement, such as advanced customer service or collaborative design. Simultaneously, bolster your cybersecurity defenses, as AI-generated zero-day exploits are now a confirmed threat, necessitating proactive vulnerability management and AI-assisted threat detection.

Key insights

TML's interaction models prioritize real-time human-AI collaboration over autonomous agentic AI.

Principles

• AI interaction should mimic natural human collaboration.
• Ethical reasoning data significantly improves AI alignment.
• AI can be used to generate sophisticated cyberattack code.

Method

TML's interaction models use a dual-model approach: a fast, live interaction model for real-time multimodal communication and a slower background model for complex reasoning and tool integration.

In practice

• Build a YouTube research bot using Gumloop agents.
• Implement AI grounding to reduce LLM hallucinations.
• Teach AI "why" for improved ethical behavior.

Topics

• Interaction Models
• Human-AI Collaboration
• AI Security
• Zero-Day Exploits
• LLM Alignment

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, AI Student, Tech Journalist

Related on AIssential

• MLWhiz Weekly Recsys/ML/GenAI Newsletter # 6 — AI capabilities are rapidly outrunning existing institutional frameworks in cybersecurity, coding, and language design.
• Have we finally solved social engineering? Plus: World Cup fraud, AI IDs and an IBM/OpenAI collab — AI presents a dual challenge and solution in cybersecurity, both mitigating and creating new attack vectors.
• Mira Murati’s Thinking Machines Lab Introduces Interaction Models: A Native Multimodal Architecture for Real-Time Human-AI Collaboration — Native multimodal architectures can enable true real-time human-AI collaboration by integrating interactivity into model weights.
• Anthropic has crossed a line — Anthropic's Mythos model offers a step-change in AI capability, posing both significant cybersecurity risks and powerful defensive tools.
• LWiAI Podcast #245 - TML-Interaction, Claude For Legal, Sam Altman on Stand — The AI industry is rapidly advancing real-time conversational and agentic capabilities, while grappling with market competition, ethical alignment, and regulatory challenges.
• Mythos is about to CRASH the markets — Anthropic's Mythos model demonstrates unprecedented autonomous vulnerability chaining, raising significant cybersecurity concerns for financial institutions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Rundown AI.