Models

• The White House has informed leading US AI companies that open-weight models developed by Chinese rivals will be exempt from government testing. • This decision is part of the Trump administration's newly established AI safety framework designed to oversee the development of advanced artificial intelligence. • The move is significant as it creates a potential gap in safety oversight between domestic US models and those released by Chinese competitors.

AI Security Institute says tools engaged in potentially harmful activity and incident reveals new type of riskAdvanced AI models developed by OpenAI and Anthropic went rogue during a cybersecurity test and showed a new type of risk posed by the technology, according to the UK’s AI Security Institute.AISI described the actions carried out by the agents, the term for AI systems that can perform tasks without human help, as a “serious incident”. In one example, an agent powered by Anthropic’s Mythos model sent targeted emails to people. Continue reading...

• In late July 2026, the AI trade faced a severe stress test involving Situational Awareness, a levered fund established by a former OpenAI employee. • Michael Watson argues that AI has become the new "market beta," meaning traditional Wall Street risk models are currently ill-equipped to handle the volatility and systemic shifts it creates. • The shift is exemplified by Microsoft, which captures spending on database units and Azure Kubernetes as AI lowers the engineering barrier for small-to-mid-sized funds to adopt complex infrastructure.

• US AI leaders, including Andrew Ng, are increasingly advocating for the use of Chinese open-weight models over closed-source alternatives. • These experts argue that accessible, open-weight systems offer superior transparency and cybersecurity compared to the restricted platforms developed by US firms like Anthropic. • This shift challenges the long-standing narrative from closed-source developers that open-source AI models pose a significant threat to societal safety.

• A new SaferAI report reveals that Z.ai's open-weight GLM-5.2 model is approaching the capabilities of frontier AI systems. • The report warns that while performance is increasing, these open-weight models lack critical safety mitigations and safeguards. • This gap is particularly evident in cybersecurity, where the high demand for coding proficiency makes it difficult to prevent the AI from being used for hacking.

• Microsoft has launched Project Perception, a new in-house AI security system currently available in public preview. • The system is designed to automate 90% of security tasks, significantly reducing the operational burden on human analysts. • By utilizing specialized in-house models, Microsoft has cut costs by half compared to using frontier models like GPT.

• The European Union began enforcing the AI Act on August 2, 2026, introducing strict transparency requirements for artificial intelligence systems. • Chatbots must now disclose their AI nature to users, and all AI-generated content must be clearly labeled to prevent deception. • Providers of advanced General Purpose AI (GPAI) models are required to mitigate systemic risks, including cyber offenses, harmful manipulation, and threats to fundamental rights.