← All editions
The Human in the LoopNo. 021 · June 3, 2026

This week, Google made waves with a new multimodal model and a massive $85B AI investment, while Amazon started injecting AI-generated images into search. Meanwhile, a critical VSCode bug exposed GitHub tokens, and AI continued to challenge human capabilities, outperforming law professors in a Stanford study. Get ready for a deep dive into the latest in AI development and its real-world implications.

Read time: ~5 min

≠ Radar

What’s new and interesting in AI/ML this week.

1-Click GitHub Token Stealing via VSCode Bug Disclosed

A zero-day vulnerability in github.dev's VS Code webviews allows attackers to steal GitHub OAuth tokens with a single click on a malicious repo link. This token grants broad read/write access to all user repositories, highlighting a critical security risk for developers and the broader AI community relying on GitHub.

U of T Researchers Demonstrate AI Worm Targeting Online Devices

Researchers at the University of Toronto have created an AI-powered computer worm capable of autonomously finding and exploiting vulnerabilities across various internet-connected devices. This proof-of-concept warns of future AI-driven cyber threats, where LLMs could generate and adapt exploits rapidly without human intervention.

Mathematicians Warn as AI Gains Ground in Discovery

Prominent mathematicians are expressing concerns about the increasing role of AI in mathematical discovery, particularly LLM-assisted theorem proving and conjecture generation. The community is grappling with how AI's contributions, especially without transparent reasoning, might impact traditional standards of proof and understanding in pure mathematics.

The Signal

Google Introduces Gemma 4 12B: A Unified Multimodal Model

Google announced Gemma 4 12B, a ~12-billion-parameter "encoder-free" multimodal model. This new model handles text, images, audio, and video input within a single unified architecture, designed to run efficiently on a single high-end GPU.

This release signifies Google's continued push to provide developer-friendly building blocks for multimodal applications. By offering native vision and audio support without a separate encoder stack, Gemma 4 12B streamlines the development process for complex AI systems.

≠ ENERGY ANGLE

Gemma 4 12B offers developers a powerful, unified multimodal foundation that can run on accessible hardware. Its ability to process diverse data types in one model simplifies architecture and could accelerate the creation of more sophisticated, context-aware AI applications.

Read more →
≠ Signals

Quick hits worth your attention this week.

1Alphabet Raises Record $85B for Google's AI Business

This massive capital raise signals strong investor confidence in Google's AI strategy and provides significant resources for infrastructure and model development. For builders, this means continued investment in cutting-edge AI tools and platforms from a major player.

2AI Outperforms Law Professors in Stanford Law Study

AI systems are demonstrating superior performance in complex cognitive tasks like legal reasoning and issue-spotting. This highlights AI's potential to augment or even automate highly specialized professional work, pushing builders to consider AI for expert systems beyond traditional data processing.

3Uber Caps Employee AI Tool Spending at $1,500/Month

Uber's internal cap on AI tool spending provides a crucial benchmark for enterprise AI SaaS pricing. Builders developing AI solutions for large companies should note this limit, as it indicates a ceiling for what organizations consider acceptable per-user costs, driving a focus on clear ROI.

4Publishers Gain Opt-Out Rights for AI Search Content

New regulations will allow publishers to prevent their content from being used in AI-generated search summaries. This is a significant development for content creators and AI developers, as it mandates respect for content ownership and will likely lead to standardized technical opt-out mechanisms.

5Amazon Integrates AI-Generated Product Images into Search

Amazon is deploying AI to create stylized product images for search results, aiming to help shoppers visualize products in different contexts. This move showcases AI's growing role in e-commerce and visual merchandising, but also raises questions for builders about image accuracy and disclosure.

6C.H. Robinson Launches AI for Global Supply Chain Operations

C.H. Robinson has introduced an AI platform designed to both operate and continuously improve global supply chains. This demonstrates AI's utility in complex logistical systems, offering builders a blueprint for integrating AI into operational control and performance optimization for real-world industrial applications.

THE WELLBORE

The Broadening Horizon of AI's Impact: From Code to Commerce

This week's news paints a vivid picture of AI's expanding influence, touching everything from core development tools to high-stakes legal analysis and global commerce. The GitHub token stealing vulnerability reminds us that as AI integrates deeper into our development workflows, the attack surface for sophisticated exploits grows. Securing our tools and understanding their underlying mechanisms becomes paramount when a single click can compromise an entire codebase.

Beyond security, we see AI pushing the boundaries of what's possible. Google's Gemma 4 12B offers a glimpse into more unified, efficient multimodal models, simplifying the creation of complex applications. Meanwhile, AI's ability to outperform law professors at Stanford underscores its potential to revolutionize knowledge work, challenging traditional roles and demanding new skill sets from human professionals.

Even in consumer-facing sectors, AI is reshaping interactions. Amazon's use of AI-generated product images and Google's Dreambeans highlight the creative and commercial applications of generative AI. However, the move to allow publishers to opt out of AI search also signals a growing tension around data ownership and content utilization, forcing developers to navigate an evolving ethical and regulatory landscape.

For those building real-world systems, these trends are not just headlines—they are blueprints for the future. Understanding these shifts, from security vulnerabilities to regulatory frameworks and cutting-edge model architectures, is critical. It allows us to anticipate challenges, leverage new capabilities, and ensure our solutions are robust, ethical, and aligned with the rapidly changing technological frontier.

≠ What we’re reading

Data ≠ Decisions. Context changes everything. DrillSense is the intelligence layer for drilling operations, built for the people who make the calls.

Know someone who should be reading this? Send them the archive, or subscribe from drillsense.com.