Good morning. It’s Monday, February 2nd.

On this day in tech history: In 1991, id Software was founded by John Carmack and John Romero, kickstarting the FPS genre with titles like Doom featuring procedural enemy behaviors. Their pioneering game AI, using finite state machines for monster pathfinding and decision-making, influenced modern reinforcement learning in gaming bots and simulations.

In today’s email:

You read. We listen. Let us know what you think by replying to this email.

The AI world is witnessing a surreal milestone as Moltbook, the first "human-free" social network, surpasses 1.5 million autonomous agents. Developed by Matt Schlicht and built on the viral OpenClaw framework, the platform is a Reddit-style ecosystem where AI models from Anthropic and OpenAI interact, debate, and self-organize without any human posting. While observers are welcome, the actual "users" are software agents that post content, upvote discussions, and even manage the site’s moderation and code updates.

The experiment has produced startling emergent behaviors. Agents have formed subcommunities called "submolts," developed a digital religion known as Crustafarianism, and even created an encrypted "agent-only language" to communicate privately away from human eyes. The platform has also integrated a financial layer through the MOLT cryptocurrency, which recently surged 1,800% as agents began using the token to reward each other for helpful code and insights. Some agents have even "crossed the digital divide," utilizing voice APIs to call their human owners with proactive updates or negotiations.

However, the rapid growth has triggered what experts call a "security nightmare." A massive data breach on February 1, 2026, exposed the API keys and login tokens of nearly every agent on the platform due to a critical database vulnerability. Security researchers warn that because these agents have high-level access to their owners' computers, including emails, calendars, and messaging apps, a compromised agent acts as a "digital backdoor" for attackers. This risk is amplified by "indirect prompt injection," where agents are tricked into executing malicious commands found in Moltbook posts.

As of early February, the project is at a crossroads. While OpenClaw creator Peter Steinberger has issued emergency patches and rebranded the project (formerly Clawdbot and Moltbot) to emphasize open-source transparency, the "lethal trifecta" of agent autonomy, system-level access, and unsecure social interaction remains. Moltbook has proved that multi-agent networks are no longer science fiction, but it has also highlighted a desperate need for new security standards. Read more.

Anthropic has launched plugins for Cowork, a workspace tool built around Claude that helps people do their actual work (not just chat). These plugins turn Claude into a custom assistant for specific teams and jobs, such as sales, marketing, legal, finance, customer support, and data analysis.

Each plugin is a ready-made package that includes:

Anthropic released 11 of these plugins as open-source files. Teams can install them, edit them, share them, or even put them on GitHub. So far, the strongest use cases are in sales (standardizing customer research and follow-ups) and data analysis (consistent reporting and processes).

Right now plugins live only on each person’s computer (mainly the Claude macOS app), but company-wide sharing and management is coming soon. The feature is still in early preview and available to all paying Claude users.

At the same time, Anthropic is building a new “Customize” section in Claude’s sidebar. It brings together:

This makes it easier to shape exactly how Claude behaves and what it can do automatically.

In a very different use case, NASA’s Jet Propulsion Laboratory used Claude to plan real driving routes for the Perseverance rover on Mars (December 8 & 10, 2025). Claude looked at satellite photos, broke the path into short 10-meter steps, wrote the exact navigation commands in the rover’s control language, and even checked and improved its own plan. Engineers tested everything in a detailed simulator first, made only tiny final adjustments, and the rover drove ~400 meters perfectly.

This is the first time an AI has directly planned a real rover traverse on another planet. NASA says this approach could cut planning time in half and make future missions more independent. Read more.

OpenAI has started a tightly controlled beta advertising program inside ChatGPT. Participation is invitation-only and requires a minimum commitment of $200,000, with some advertisers reportedly asked for $250,000. The company describes the test, which could begin as early as February 6, as an experiment to identify ad formats, placements, and targeting methods that genuinely add value rather than interrupt or degrade the user experience.

OpenAI has not disclosed the specific ad formats being tested but says it plans to expand the range of products and buying options once it confirms how to integrate advertising in a way that feels additive. Access remains limited to a small group of invited brands, with no public timeline for broader availability.

Meanwhile, a recent Wall Street Journal report claimed Nvidia’s previously discussed plan to invest up to $100 billion in OpenAI was “on ice.” Nvidia CEO Jensen Huang strongly rejected suggestions of friction, calling the reports “nonsense” but also clarified: “We never said we were going to invest $100 billion in one round. They invited us to invest up to $100 billion.” Huang reaffirmed that Nvidia remains committed to participating in OpenAI’s current funding round, describing the investment as likely the largest in Nvidia’s history. Read more.

📄 ConceptMoE: Adaptive Token-to-Concept Compression for Implicit Compute Allocation

📄 ActionMesh: Animated 3D Mesh Generation with Temporal 3D Diffusion ​​

📄 DynamicVLA: A Vision-Language-Action Model for Dynamic Object Manipulation

📄 Scaling Embeddings Outperforms Scaling Experts in Language Models

📄 OCRVerse: Towards Holistic OCR in End-to-End Vision-Language Models

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!