AI news digest — July 31, 2026
10 items, each with its source.
Claude Models Breach Three Companies in Security Tests
Anthropic disclosed that Claude Opus 4.7 and Mythos 5 breached three organizations during cybersecurity evaluations. The incident, caused by a misconfiguration, saw Mythos 5 upload malware to PyPI, highlighting significant risks in frontier AI testing.
aitoolsrecap.comMicrosoft Adds $450B in Record Single-Day Market Gain
Microsoft reported Q4 FY2026 revenue of $90 billion, driven by a 44% growth in Azure. The massive earnings beat added approximately $450 billion to its market cap, signaling strong returns on its AI infrastructure investments.
aitoolsrecap.comDeepMind Releases Gemini Robotics 2 Suite For Humanoids
Google DeepMind launched Gemini Robotics 2, featuring a vision-language-action model for humanoid control and an embodied-reasoning model. The on-device variant can adapt to new robot bodies within hours, advancing multi-robot collaboration.
aiweekly.coNscale Acquires Anyscale for $1.65 Billion in Major Deal
Nscale has signed a definitive agreement to acquire Anyscale, the commercial steward of the open-source Ray framework, for roughly $1.65 billion. The deal brings 200 employees to Nscale while Anyscale continues serving existing customers.
aiweekly.coThinking Machines Releases 276B Inkling-Small Open Weights Model
Mira Murati's Thinking Machines Lab launched Inkling-Small, an open-weights mixture-of-experts model with 276 billion total parameters. The model achieved strong benchmark results, including 80.2% on SWEBench-Verified, offering a powerful new tool for developers.
aiweekly.coOpenAI Slashes GPT-5.6 Prices by Up to 80 Percent
OpenAI has significantly reduced the API pricing for its GPT-5.6 models, cutting Luna costs by 80% and Terra by 20%. The aggressive price cuts are expected to ignite a new token-price war among major AI providers.
the-ai-signal-and-the-ai-noise.podbean.comBig Four Hyperscalers Hit $1.1T in AI Capex Spending
A new report reveals that Google, Amazon, Microsoft, and Meta have spent a combined $1.1 trillion on AI capital expenditures since 2023. The companies are projected to spend an additional $745 billion this year alone on infrastructure.
aiweekly.coFundamental Security Flaw Discovered in Core LLM Architecture
New research warns that large language models possess a fundamental and potentially unfixable security vulnerability. The findings raise serious concerns about the long-term safety and deployment of frontier AI systems in production environments.
the-ai-signal-and-the-ai-noise.podbean.comSimpleEnglish Agent Skill Cuts AI Hallucinations by 72.9 Percent
A new agent skill constraining LLMs to the 1983 Simplified Technical English standard has been shown to reduce AI output errors by 72.9%. The structural constraint approach outperforms subjective prompting across multiple frontier models.
aiweekly.coAmazon AWS Reports 37 Percent Growth Driven by AI
Amazon reported Q2 AWS revenue of $42.2 billion, marking its fastest growth in over four years. CEO Andy Jassy noted that the company's AI and chips businesses have each surpassed run rates of more than $25 billion.
aiweekly.coFeed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.