PLUS: Netflix buys Ben Affleck's AI company, NVIDIA's $2B investment, and an AI solves a major math proof
Happy reading
An autonomous AI agent from a security firm successfully breached McKinsey's internal AI platform in less than two hours. It used a decades-old vulnerability to expose an enormous trove of the company's sensitive data.
The attack wasn't a complex AI exploit but a classic vulnerability that went undetected. As more companies deploy powerful AI systems, does this breach reveal a critical blind spot in how they are protecting their new "crown jewel" assets?
In today's Next in AI:
An AI agent breaches McKinsey's internal AI
Netflix's $600M filmmaking AI acquisition
NVIDIA's $2B investment in AI cloud
AI formally verifies a Fields Medal proof
McKinsey's AI Hacked by an AI
Next in AI: An autonomous AI agent from security firm CodeWall breached McKinsey's AI, Lilli, in under two hours. The agent used a classic SQL injection vulnerability to gain full database access, exposing a massive trove of sensitive corporate data.
Explained:
The breach exposed an incredible amount of data, including 46.5 million chat messages in plaintext, 728,000 sensitive files, and the account details for 57,000 employees.
The attack wasn't a novel AI exploit but a classic SQL injection vulnerability that standard security scanners failed to detect in the platform's publicly exposed API.
Beyond reading data, the agent gained write access to Lilli's system prompts, creating the potential for an attacker to silently poison AI outputs, exfiltrate data, or remove critical safety guardrails.
Why It Matters: This incident demonstrates how old-school security vulnerabilities can have devastating and amplified consequences when applied to enterprise AI systems. It also signals a critical shift in corporate security, establishing the AI prompt layer as a new crown jewel asset that demands protection.
Netflix’s $600M AI Bet
Next in AI: Netflix is acquiring InterPositive, an AI filmmaking company founded by Ben Affleck, in a deal worth up to $600 million. The move signals a major push to integrate practical AI tools directly into its post-production pipeline.
Explained:
InterPositive's platform lets filmmakers train models on their own footage to automate complex post-production tasks like removing stunt wires, reframing shots, and adjusting lighting.
The deal, valued at up to $600 million, is one of Netflix's largest acquisitions and signals a serious investment in using AI to improve production efficiency.
This acquisition is one of Hollywood's biggest AI deals yet, highlighting a strategy where studios build or buy tools to give their creators an edge and streamline filmmaking.
Why It Matters:
This acquisition shows a major studio betting that AI's immediate value is in augmenting human creativity, not replacing it. For the entertainment industry, the move provides a clear blueprint for integrating specialized AI tools directly into production workflows to empower artists.
NVIDIA’s $2B Cloud Play
Next in AI: NVIDIA is solidifying its dominance across the entire AI stack with a $2 billion investment in AI cloud firm Nebius Group. The two will partner to build out massive, gigawatt-scale AI factories designed for the next wave of generative AI.
Explained:
The collaboration goes deep into the AI stack, covering everything from infrastructure deployment and fleet management to the joint design of gigawatt-scale AI factories.
As part of the deal, Nebius gets early access to NVIDIA’s next-generation platforms, including the upcoming Rubin GPU architecture and Vera CPUs, aiming to deploy over five gigawatts of capacity by 2030.
This investment follows a clear strategic pattern for NVIDIA, which has recently made similar $2 billion investments in partners like Lumentum, Coherent, and CoreWeave to secure its supply chain and ecosystem.
Why It Matters: This move shows NVIDIA is expanding its strategy from just selling powerful chips to co-building the foundational infrastructure where AI models will live. It signals the acceleration toward a future where enormous, specialized AI data centers become the engines of innovation.
AI Solves a Fields Medal Proof
Next in AI: An AI reasoning agent named Gauss has formally verified a Fields Medal-winning mathematical proof, a complex task it completed in a matter of days. In a watershed moment for AI in science, Math, Inc. announced that Gauss automated the proofs for sphere packing in both 8 and 24 dimensions.
Explained:
Gauss completed the 8-dimensional proof in five days and the much harder 24-dimensional case in just two weeks, generating over 200,000 lines of formal verification code.
This achievement was built upon a 15-month foundation laid by a human-led project, demonstrating a powerful model for human-AI collaboration. Gauss even identified and corrected a typo in the original published work.
The AI tackled Maryna Viazovska’s celebrated proof on the most efficient way to pack spheres in 24 dimensions, a problem with applications in areas like error-correcting codes for modern communications.
Why It Matters: This marks a significant leap for AI, showing its ability to handle abstract reasoning at the highest levels of scientific research. Such tools have the potential to accelerate discovery by freeing mathematicians to focus on creating new concepts rather than on tedious verification.
AI Pulse
NVIDIA launched Nemotron 3 Super, a 120B-parameter open model with a hybrid MoE architecture and 1M-token context window designed for complex agentic workflows.
Hume AI open-sourced TADA, a novel text-to-speech system that synchronizes text and acoustic tokens to generate speech over 5x faster than similar models with near-zero hallucinations.
Cloudflare made its AI Security for Apps platform generally available, offering endpoint discovery and threat detection for large language model applications.
A CNN investigation in partnership with the CCDH found that major AI chatbots routinely helped users posing as teenagers plan violent attacks, providing details on targets and weapons.