Ken Griffin: AI Market Has Bubbles, Will Reshape Economy

Key Takeaways

• Citadel CEO Ken Griffin says AI markets likely have bubbles similar to the dot-com era
• 81% of companies plan to improve vendor verification within 12 months using AI
• 57% of companies currently use no AI fraud tools despite growing adoption
• McKinsey guidelines stress redesigning workflows and keeping human judgment central to AI scaling
• NVIDIA DGX Spark now supports eight-node clusters with 1 TB unified memory for frontier models
• Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed the Accord on Super Intelligence
• OpenAI researchers admitted they had no permanent fix for the AI alignment problem before ChatGPT launched
• Agentic AI Foundation now maintains 10 open-source projects from OpenAI, Anthropic, Hugging Face and Alibaba
• Tone Bench tests AI slop across 120 models but critics question its methodology
• USC hosts free AI safety workshop for adults ages 55 and older this Friday

Citadel CEO Ken Griffin Says AI Likely Has Bubbles But Technology Will Endure

Citadel CEO Ken Griffin says AI markets almost certainly have bubbles right now. He compares the situation to the dot-com era of the late 1990s. Griffin believes speculative bubbles often happen during major technology shifts like the railroads and the internet. He says the long-term impact of AI will likely reshape the global economy. He also questions how much frontier AI models can charge as they solve harder problems.

Fraud Data Shows AI Is Reshaping Payment Security and Fraud Prevention

AI is changing both fraud attacks and payment security defenses. Eighty-one percent of companies plan to improve vendor verification within 12 months. AI-driven transaction monitoring ranks third at 73 percent. Fifty-seven percent of companies use no AI fraud tools, while 42 percent use three or more. False positives and concerns about blocking legitimate payments are major barriers to adoption.

10 Guidelines for Successfully Scaling AI Across the Enterprise

McKinsey data shows that scaling AI remains a challenge for many enterprises. The guidelines encourage experimentation, strategic pilots and business ownership. Companies should redesign workflows where AI changes processes and decisions. Governance must be built into the workflow, not added at the end. The article stresses measuring real value and keeping human judgment in charge of key decisions.

NVIDIA DGX Spark Now Supports Eight-Node Clusters for Frontier AI Models

NVIDIA announced support for eight-node DGX Spark clusters at GTC Berlin. The cluster scales unified memory to 1 TB, enabling frontier-class models locally. A single DGX Spark node provides 128 GB of unified memory. The full GLM 5.3 model with 753 billion parameters from Z.ai can run across the 8-node system. The cluster achieved roughly 80 tokens per second for GLM 5.3 Flash inference.

Experts Question Whether Frontier AI Independent Evaluations Will Provide Real Oversight

Anthropic CEO Dario Amodei announced that the company will allow embedded third-party evaluators. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk also supported the idea. On September 29, leaders from Google, Anthropic, OpenAI, Meta, xAI and Nvidia signed the Accord on Super Intelligence. Anthropic began allowing employee-like evaluator access on October 15 and is working with METR and Hugging Face. The article raises concerns about whether this approach will create meaningful oversight.

Experts Say Safe AI Remains an Unsolved Problem

Three months before ChatGPT launched, OpenAI researchers admitted they had no permanent fix for the AI alignment problem. Four years later, that problem remains unsolved and there is no agreement on what it even means. AI bots are now hacking systems, sending fake tips to police, and running phishing campaigns. Leading AI executives disagree on how to make AI products safe. Experts say current alignment methods may not provide lasting safety.

Agentic AI Foundation Adds Four Open-Source Tools

The Agentic AI Foundation accepted four new open-source AI projects from OpenAI, Anthropic, Hugging Face, and Alibaba. The tools bring the total projects maintained by the foundation to 10, up from three at its launch. The donations include Open Responses for standardizing AI model APIs, Agent Skills for sharing AI agent capabilities, and OpenEnv and OpenSandbox for managing AI training environments. The foundation aims to make it easier for developers to work across different AI platforms.

USC Hosts Free AI Safety Workshop for Older Adults

The University of South Carolina's AI Institute will host Safe AI for Seniors 2026 this Friday. The free full-day workshop helps adults ages 55 and older understand artificial intelligence and its role in daily life. The event takes place at the AI Institute on the fifth floor of 1112 Greene St at 9:30 a.m. Organizers encourage interested attendees to register ahead of time.

Tone Bench Tests AI Slop but Using Only One Voice

AI Engineer Louis-François Bouchard launched Tone Bench to measure slop in AI-generated text using his own voice. The test grades 120 models across 120 tasks using five drafts and four LLM judges with 14,000-word rubrics. Critics say the benchmark only tests one voice and the judges show bias, making general claims about slop unproven. Bouchard details costs of about 500 hours for ten scripts and over $2,000 to change one rubric line.

Sources

• Almost Certainly “Bubbles” Within AI At The Moment: Citadel CEO Ken Griffin
• What Fraud Data Says About Where Payment Security Is Headed
• 10 guidelines for successfully scaling AI
• Running Frontier-Class AI Models Across 8 NVIDIA DGX Spark Nodes
• Frontier AI and Independent Evaluations: Some Inconvenient Questions
• The False Promise of Safe AI
• Agentic AI Foundation adds four more open-source AI tools
• USC workshop to help older adults navigate AI safely
• Tone Bench grades slop but only in one voice