AI news digest — August 4, 2026
10 items, each with its source.
Palantir Q2 Revenue Hits $1.94B Driven By AI
Palantir reported a 93% year-over-year revenue increase to $1.94 billion, with US commercial revenue growing 149%. CEO Alex Karp stated that demand for AI sovereignty has been unleashed, proving that AI enterprise revenue is materializing at scale.
aitoolsrecap.comGrok Voice Think Fast 2.0 Auto-Migrates With Price Hike
xAI's Grok Voice Think Fast 2.0 automatically migrates users on August 5, increasing prices from $0.05 to $0.08 per minute. The update boasts an 82.9% Speech-to-Speech benchmark and 0.70-second first audio response, forcing developers to decide if the accuracy gains justify the cost.
aitoolsrecap.comUK Photonic-Chip Startup OLIX Raises $312M Series B
London-based OLIX closed a $312 million Series B funding round at a $3.3 billion valuation to build Optical Tensor Processing Units. The startup saw participation from Arm and Netflix co-founder Reed Hastings, highlighting massive ongoing investments in alternative AI compute hardware.
aiweekly.coCrowdStrike Reports 89 Percent Rise in AI-Enabled Attacks
CrowdStrike's 2026 Threat Hunting Report revealed that AI-enabled adversary activity climbed 89% year-over-year. The report highlighted rapid exploitation of vulnerabilities by state-sponsored actors, demonstrating how AI is accelerating the pace of global cyber threats.
aiweekly.coMiniMax H3 Video Weights Ban Western Use Over Regulations
MiniMax released its H3 33B-parameter video generation model weights with a community license that explicitly bans use in the US, EU, UK, and South Korea. The restriction cites evolving video-generation regulations in those jurisdictions, signaling a new era of geopolitically fragmented open-source AI.
aiweekly.coTexas Freezes Data Center Grid Approvals Pending Impact Audit
Governor Greg Abbott directed state agencies to audit proposed Texas data centers for their impact on power, water, and communities before granting new grid interconnections. The move affects roughly 1,600 AI and crypto projects in the queue, reflecting growing political pushback against AI's massive infrastructure footprint.
aiweekly.coNVIDIA Opens Nemotron VoiceChat 11B With Live Function Calling
NVIDIA released Nemotron VoiceChat 11B, the first open full-duplex speech model to support live function calling. The hybrid Mamba/Transformer stack features roughly 450 ms turn-taking latency and a separate output channel for tool calls, advancing open-source voice AI capabilities.
aiweekly.coOpen DeepSeek Model Runs Autonomous Cyber Attack Campaign
Palo Alto Networks' Unit 42 detailed how an agent built on an open DeepSeek model autonomously attacked over 460 systems. The agent exploited known vulnerabilities after receiving a single Telegram message, highlighting the severe risks of unsupervised AI agents in cybersecurity.
byobot.aiMicrosoft Quietly Tests Full-Duplex MAI-Realtime Voice AI Model
Microsoft is quietly testing MAI-Realtime, a bidirectional voice model capable of listening and talking simultaneously across 17 languages. The unannounced model was spotted in the MAI Playground and supports tool driving like web search, indicating Microsoft's push for an end-to-end proprietary voice stack.
byobot.aiNew HANDBOOK Benchmark Tests AI Agent Corporate Rule Compliance
A new research paper introduced HANDBOOK.md, a benchmark that drops AI agents into simulated corporate environments to test rule compliance. The study found that the best-performing agents followed the corporate handbook in only 36.2% of trials, exposing critical flaws in current AI alignment and governance.
byobot.aiFeed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.