PLUS: the first AI-generated zero-day, Google's new AI pointer, and why AI layoffs are failing
Happy reading
OpenAI is open-sourcing a key piece of its infrastructure for building massive AI models. The new network protocol, developed with partners, is designed to speed up training and make the supercomputers that power AI more efficient.
By releasing the blueprints for this high-performance networking technology, OpenAI is helping to lower the barrier for others to build at scale. But will this move truly democratize the race to AGI, or are there still too many other hurdles for smaller players to overcome?
In today’s Next in AI:
OpenAI open-sources its AI supercomputer protocol
Google’s new AI-powered mouse pointer
The first confirmed AI-generated zero-day
Why AI-driven layoffs are failing to deliver ROI
The Network for AGI
Next in AI: OpenAI, with partners including Microsoft and NVIDIA, is open-sourcing a new protocol to accelerate training for massive AI models. They released the specification for MRC (Multipath Reliable Connection) to help the entire industry build more resilient and efficient AI supercomputers.
Explained:
Instead of sending data down a single path, MRC sprays packets across hundreds of routes at once, virtually eliminating network traffic jams that can idle expensive GPUs.
The protocol uses source routing to bypass network failures in microseconds, allowing training to continue uninterrupted even when switches are rebooted or links fail.
This design enables building massive supercomputers with over 100,000 GPUs using fewer tiers of switches, reducing power consumption and overall complexity.
Why It Matters: This move helps democratize the high-performance networking needed for frontier AI, enabling more organizations to build at massive scale. As models grow, this kind of foundational infrastructure becomes essential for making progress toward more capable AI systems.
The AI Mouse
Next in AI: Google is reimagining the mouse pointer for the AI era. The humble cursor is becoming a context-aware tool that lets you interact with your screen using simple spoken commands and gestures.
Explained:
Instead of requiring detailed text prompts, the AI-powered pointer captures visual and semantic context, allowing for natural shorthand like pointing at an object and saying, "summarize this."
The technology is being integrated into existing products, appearing in Gemini in Chrome and as a "Magic Pointer" feature on Google's new Googlebook laptop.
This approach is guided by principles that turn pixels into actionable entities, transforming a static image of a restaurant into a booking link or a data table into a chart, as shown in experimental demos.
Why It Matters: This development pushes AI out of separate chat windows and directly into our core user interfaces, reducing the friction between intent and action. It signals a future where computers adapt to how humans communicate, rather than forcing us to learn the language of machines.
The AI Zero-Day
Next in AI: Google's Threat Intelligence group has confirmed the first-ever case of hackers using AI to help discover a zero-day vulnerability capable of bypassing two-factor authentication. The official report details how AI is being weaponized for offensive cybersecurity.
Explained:
The attack script showed clear signs of AI generation, including overly detailed comments, textbook formatting, and a hallucinated security score, which human attackers would typically omit.
AI models are proving particularly effective at spotting high-level logic flaws in code—the kind of semantic errors that traditional security scanners often miss.
This isn't an isolated case; Google also noted that state-sponsored groups are using AI to rapidly analyze vulnerabilities and test exploits at a scale previously requiring immense human effort.
Why It Matters: The theoretical threat of AI in cybercrime is now a confirmed reality, kicking off a new arms race between attackers and defenders. This discovery pressures security teams to adopt their own AI tools to find and patch complex vulnerabilities before they are exploited.
AI's ROI Problem
Next in AI: A new Gartner study finds that companies laying off workers for AI are not seeing a return on investment. The highest gains are found in firms using AI to amplify their existing workforce, not replace it.
Explained:
A survey of 350 global executives revealed that while 80% of firms piloting AI also reported workforce reductions, these cuts showed no correlation to higher ROI.
Instead, the highest returns come from practicing ‘people amplification,’ a strategy that helps top-performing firms capture the vast majority of AI-driven value.
This insight comes as AI is cited as a leading reason for layoffs, with one report tracking nearly 50,000 job cuts in the tech sector so far this year.
Why It Matters: The data suggests a major strategy shift is needed for companies investing in AI. True value lies in augmenting human capabilities, not just chasing headcount reductions.
AI Pulse
Cerebras upsized its IPO to a target of $4.8 billion, a deal that includes granting OpenAI warrants for up to a 10% stake in the company.
Gigacatalyst launched an embedded AI builder that allows non-technical users to create custom workflows and features for SaaS products using natural language.
Graduates booed their University of Central Florida commencement speaker for calling AI "the next Industrial Revolution," highlighting growing anxiety over AI's impact on creative jobs.
OpenAI weighed in on the trend of developers keeping laptops open in public to let coding agents run, a new habit emerging from the push for longer, uninterrupted AI coding sessions.