The morning briefing.

Frontier AI Safety. Anthropic's Claude Mythos has become the first AI model to pass all cyberattack simulations by Britain's AI Security Institute, marking a significant milestone in AI safety. This achievement comes as the UK agency revised its estimate for AI cyber capabilities doubling, highlighting rapid advancements. However, Anthropic's head of red teaming warns that even Mythos will likely seem "quite dumb" within a year, underscoring the fast-evolving nature of AI.

Corporate Strategy Shifts. Cisco announced plans to lay off 4,000 workers, or about 8% of its workforce, despite a surge in AI-related orders. CEO Chuck Robbins stated these are "clear, strategic investments" as the company shifts focus to high-growth areas. Meanwhile, Chinese tech companies are accelerating the development of homegrown AI chips, even as the US reportedly allows ten Chinese firms to purchase NVIDIA's H200 AI chips, signaling a complex and evolving global semiconductor landscape.

Digital Security Evolution. The cybersecurity landscape continues to evolve with the arrest of a man suspected of being the Dream Market kingpin, facing charges in both the US and Germany. Simultaneously, a new Linux kernel flaw, "Fragnesia," has emerged with public exploit code, posing a significant privilege escalation risk. These developments highlight the ongoing battle against sophisticated cyber threats and the critical need for robust security measures across operating systems and digital platforms.

AI Implementation Hurdles. Many organizations are already rolling back AI customer service tools, indicating that advanced firms are encountering failures sooner rather than less often. A new study reveals that employees are spending a full day each week managing disconnected AI systems, acting as "human middleware." This suggests that current AI deployments often create more complexity than they solve, highlighting the need for better integration and more thoughtful implementation strategies to avoid costly infrastructure traps.

Advanced Hardware and Compute. NASA is developing a next-generation processor that is 500 times more powerful than current space chips, enabling spacecraft to make autonomous decisions. This innovation aligns with broader trends in hardware, including the acceleration of chipmaking for energy-efficient AI. Meanwhile, a Prague-based startup, Zerops, raised €1.7 million to enhance cloud infrastructure by bridging the gap between development and production, supporting global expansion into the US and Asia.