The full desk, newest first.
Every story Pulse24 is tracking, in chronological order — newest first.
iroh Launches Mesh LLM for AI
iroh's Mesh LLM pools existing GPUs for distributed LLM inference via an OpenAI-compatible API. This shifts compute from metered cloud services to owned infrastructure, giving teams greater control over execution and reducing operational costs.
Hyperscalers Fund Neocloud GPU Expansion
Hyperscalers Microsoft and Meta committed over $122.2 billion to neoclouds CoreWeave and Nebius, securing rapid access to frontier GPU compute. This shifts significant capital expenditure and accelerates AI infrastructure deployment, impacting procurement and platform engineering teams.
Anthropic Reveals Claude's Internal Reasoning
Anthropic's new "Jacobian Lens" technique reveals Claude's internal reasoning, offering a framework for safer, more transparent AI. This mechanism allows auditing model behaviour, potentially identifying hidden goals before they manifest in outputs, addressing the "black box" problem.
Apple Sues OpenAI Over Trade Secrets
Apple's lawsuit against OpenAI for trade secret theft complicates OpenAI's IPO plans and shifts risk for investors. The legal action, alleging misappropriation via a former Apple VP, highlights increasing intellectual property scrutiny in the AI sector for founders and procurement teams.
Air Force Accelerates Anthropic Purge
Air Force contractors must remove Anthropic products by September 1, nearly a month before the Department of Defense's wider deadline. This accelerates the timeline for identifying and replacing AI dependencies, creating immediate operational and compliance challenges for defence suppliers.
Meta Begins Iris Chip Production
Meta Platforms will begin mass production of its in-house AI chip, "Iris," as early as September. This move aims to reduce reliance on external suppliers and supports a $125B-$145B capital expenditure for AI infrastructure, impacting procurement and CTOs.
Meta AI Detector Fails Cropped Images
Meta's new AI image detection tool failed to identify 55% of its own AI-generated images after cropping, exposing a critical vulnerability in watermark-based provenance systems. This mechanism failure increases deepfake risk, requiring security architects and content moderation teams to assume current watermark technology is not fully robust.
Media Outlets Demand OpenAI Sanctions
Legal teams face heightened scrutiny over AI training data provenance as media outlets seek sanctions against OpenAI for alleged hidden evidence of copyrighted training data. This action could compel disclosure of datasets and logs, setting a precedent for AI model development transparency.
Xbox CEO Joins Fed AI Taskforce
Xbox CEO Asha Sharma's appointment to the US Federal Reserve's AI taskforce means economic policy will directly incorporate tech executive insights. Founders and investors must track these emerging policy judgments, which will shape the regulatory and economic landscape for AI development.
Lyzr Agent Secures $100M Series B
AI agents are now directly executing high-stakes financial operations. Lyzr's SivaClaw autonomously managed its $100 million Series B fundraise, securing $400 million in investor interest and a $500 million valuation without founder travel, demonstrating reduced fundraising overhead.
Mac Mini Powers Agentic AI
Apple's Mac mini and Mac Studio are seeing high demand for agentic AI workloads, driven by Apple silicon's power-efficient Neural Engine and unified memory. This offers a budget-friendly, secure option for local AI agent deployment, reducing reliance on costly cloud tokens.
OpenAI GPT 5.6 Powers Microsoft Copilot
OpenAI's GPT 5.6 is now the "preferred model" for Microsoft 365 Copilot, indicating continued external model use despite Microsoft's internal AI development. This suggests Microsoft's cost-reduction strategy via internal models is not yet fully implemented for Copilot's core, necessitating hybrid model strategies for platform engineers.
OpenAI Integrates AI Browsing Features
OpenAI is discontinuing its Atlas browser, integrating its AI agent capabilities into the ChatGPT desktop app and a new Chrome extension. This centralises AI functionality within existing workflows, altering how users engage with web content through AI.
Meta Launches Muse Spark 1.1 API
Meta's new Muse Spark 1.1 model, designed for agentic coding, is now available via a paid API with competitive token pricing. This adds a competitive option for enterprise AI automation, directly impacting procurement teams evaluating solutions for large-scale coding tasks and workflow management.
Brown Professor Reports Midterm Score Surge
Unproctored assessments, when combined with AI assistance, can lead to inflated academic performance, potentially masking a lack of actual learning, which risks undermining academic integrity and the validity of student evaluations.
Databricks Benchmarks Coding Agents
Engineering teams must adopt a multi-model, multi-harness strategy for coding agents. Databricks' benchmark shows open models like GLM 5.2 offer top quality at lower costs, while harness choice can cut expenses by over 2x, shifting optimal deployment strategies.
China Plans Limited Nvidia H200 Sales
Chinese AI firms may soon access limited Nvidia H200 chips, easing a critical computing capacity crunch. This controlled approval, potentially under 200,000 units, signals a strategic shift in China's hardware procurement.
Solidigm Elevates Storage for Agentic AI
GPU underutilisation due to data bottlenecks now drives significant AI infrastructure costs. Solidigm positions high-capacity SSDs as the "intelligence layer" for agentic AI inference, shifting storage from commodity to critical for continuous GPU utilisation and efficient token generation.
xAI Grok 4.5 Leads Coding Performance
xAI's Grok 4.5 model offers significantly lower costs and higher throughput for coding tasks compared to GPT-5.5 and Claude models. This shifts unit economics for high-volume code generation, providing a new option for platform engineers and CTOs.
NHTSA Orders AVs Stop Blocking
NHTSA demands autonomous vehicle developers immediately resolve AV interference with first responders, citing a "functional insufficiency." Product teams must prioritise emergency scene detection, facing potential regulatory action and accountability if solutions are not presented by month-end.