John Ternus Takes Over as Apple CEO — AI Is Job Number One
Today marks a historic transition at the world's most valuable company. John Ternus, Apple's long-time hardware engineering chief, has officially taken the reins as CEO, with Tim Cook moving to the role of Executive Chairman. Ternus, who has spent 25 years at Apple working on Macs, iPads, AirPods, and iPhones, becomes the company's eighth chief executive.
The timing couldn't be more significant. Apple has faced mounting criticism for lagging behind competitors in AI, and closing that gap is now Ternus's top mandate. According to Bloomberg News reporting, his immediate priorities include a rebuilt Siri powered by large language models and an ambitious product pipeline featuring the company's first foldable iPhone, camera-equipped AirPods, smart glasses, and a robotic-armed home display — all central to Apple's AI-era strategy.
An Apple event is already set for September 9, where Ternus is expected to unveil the biggest-ever change to the iPhone lineup. With COO Sabih Khan, CFO Kevan Parekh, and Services Chief Eddy Cue taking on expanded roles, the new leadership team is positioning Apple to compete more aggressively in the AI race.
Anthropic Shows Self-Improving AI Can Fix Alignment Problems
In what may be one of the most consequential AI safety developments of the year, Anthropic published a landmark paper titled "Automated Researchers Can Reliably Mitigate Alignment Failures", demonstrating that AI systems can improve other AI models' safety with minimal human involvement.
Led by researcher Chen Yueh-Han, the team gave Claude Sonnet 5 an early version of the more powerful Claude Opus 4.8 and tasked it with making the model behave better. Over roughly 60 hours, the automated system tested more than 50 different ideas before creating a training method using about 2,400 examples. When given 10 benchmarks for specific misaligned behaviors, the automated systems improved performance on every single one without degrading overall capability.
The economics are striking: the automated researchers completed tasks at roughly $4 per hour in API inference costs, compared to $150 per hour for human researchers — and they were faster too. As TechCrunch noted, this represents an early but real glimpse at self-improving AI — a concept that until recently lived primarily in the realm of theoretical concern. MIT Technology Review offered a counterpoint, noting that truly recursive self-improvement may still be further off than these results suggest.
DeepSeek Closes In on $74 Billion Valuation Ahead of IPO
Chinese AI powerhouse DeepSeek is nearing completion of a massive $7.4 billion funding round at a pre-money valuation of approximately $74 billion, according to Bloomberg. This comes just months after the company closed a $7 billion round in May at a $52 billion valuation — a pace of fundraising that underscores the breakneck capital intensity of frontier AI development.
Existing backers Monolith, Shixiang Capital, and CATL are joined by new investors including CPE, Legend Capital, and Stony Creek Capital, along with funds backed by GigaDevice and Chinese state investment vehicles. The proceeds will fund roughly 1 gigawatt of new compute capacity, as DeepSeek's efforts to develop AI agents drive significantly greater demand for computing power.
The company, founded by Liang Wenfeng through his hedge fund High-Flyer, is targeting an IPO filing as early as late 2026 on Shanghai's Star Market, with a public debut expected in 2027. The fundraising intensifies talent and infrastructure competition against Alibaba's Qwen, Tencent's Hy4, and Zhipu's GLM.
McKinsey's State of AI 2026: The Hype-Reality Gap Widens
McKinsey's "State of AI" Global Survey 2026 paints a sobering picture of enterprise AI adoption. While 62% of organizations say they are at least experimenting with AI agents, only 23% are scaling an agentic system anywhere in the enterprise — and in any single function, the ceiling is 10% or less.
Perhaps the most telling finding: 32% of organizations have skipped purchasing at least one software product or feature because they could build it internally with agentic coding tools. This marks a significant shift in how companies view the build-vs-buy equation in the age of AI-assisted development.
Yet the ROI gap persists. Nearly two-thirds of organizations have not begun scaling AI across the enterprise, and only 39% report EBIT impact at the enterprise level despite widespread use-case-level deployments. Only 25% of leaders expect AI agents to act as autonomous teammates in the short term — most still see AI primarily as a support tool for the next one to two years.
Quick Hits
Anthropic locks in Claude Sonnet 5 pricing: Anthropic confirmed that Claude Sonnet 5's introductory pricing ($2/M input, $10/M output tokens) is now permanent, canceling the planned September 1 increase to $3/$15. Note that Sonnet 5's new tokenizer counts roughly 30% more tokens per equivalent text.
Meta's Muse Glimmer runs AI agents on your laptop: Meta's 30B-parameter Muse Glimmer runs offline on a single consumer GPU to execute local coding and autonomous agent workflows. Released under Apache 2.0, 4-bit quantization compresses it to 18–20 GB, fitting within a 24 GB VRAM envelope.
US states crack down on AI chatbots: Nearly 100 chatbot-specific bills have been introduced across 34 states, with California, Oregon, Connecticut, and Washington leading in passed legislation — particularly around companion chatbot transparency and mental health safeguards for minors.