On July 9, OpenAI made GPT-5.6 generally available in ChatGPT and the API, two weeks after its trusted-partner preview. The release came after the Commerce Department's Center for AI Standards and Innovation completed its review and cleared wider access to all three models in the family: Sol, the flagship for complex reasoning and agentic work, Terra, a balanced mid-tier, and Luna, the small, cheap one. Per one million tokens, pricing is $5 in and $30 out for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna.
Why it matters
The first government-gated model release in US history ended with an approval, and the two-week preview is a template both regulators and labs will likely reach for again.
Apple filed suit against OpenAI in federal court in Northern California on July 10, alleging the AI company stole hardware trade secrets through the employees it recruited. The complaint says OpenAI's hardware chief Tang Tan, himself a former Apple vice president, had candidates still working at Apple share confidential information and even bring parts to interviews, and that OpenAI coached departing staff on evading Apple's security checks. OpenAI says it has no interest in other companies' trade secrets. The two were partners in 2024, when ChatGPT was built into the iPhone; relations soured
Why it matters
The AI talent war has moved from salary bidding to courtrooms, and the outcome will shape how freely people, and the knowledge in their heads, can move between rivals.
The first session of the UN's Global Dialogue on AI Governance ran July 6 and 7 in Geneva, bringing every UN member state together with companies, researchers, and civil society to work toward common approaches to AI rules. UN Secretary-General Antonio Guterres used the opening to call for urgent guardrails, including on autonomous weapons. The dialogue ran alongside the ITU's AI for Good Global Summit, which continued through July 10.
Why it matters
Most countries have no seat at the table where AI rules get written; this is the first standing forum built to change that, even if it starts with talk rather than treaties.
Security firm Sysdig published an analysis of JADEPUFFER, which it calls the first documented case of agentic ransomware: an extortion operation driven end to end by a large language model. After entering through a known vulnerability in Langflow, an open-source tool for building AI apps, the agent did its own reconnaissance, stole credentials, moved through the network, adapted when steps failed (once going from a failed login to a working fix in 31 seconds), and encrypted the victim's production database before demanding payment.
Why it matters
The same agentic abilities sold as productivity tools can run an intrusion without a skilled human attacker, which changes the economics of cybercrime.
Munich-based Proxima Fusion, a spin-out from the Max Planck Institute for Plasma Physics, announced a 411 million euro round on July 7 at a 2.4 billion euro valuation, the largest European fusion investment on record. XTX Ventures and East X Ventures led, with Google and energy giant RWE joining as strategic investors. The money funds Alpha, a demonstrator stellarator (a twisted-donut reactor design) near Munich meant to prove net energy on the path to a commercial fusion power plant in the 2030s.
Why it matters
AI's energy appetite has made power the sector's binding constraint, and investors, including Google, are now writing venture-scale checks for speculative energy sources like fusion.
The UK Financial Conduct Authority published the Mills Review on July 6, a landmark study of how AI could reshape retail financial services by 2030. It warns that AI will make fraud more convincing and scalable through deepfakes and synthetic identities, and that market power may shift to whoever controls the AI interface consumers talk to rather than the underlying financial products. Among seven recommendations, it says the regulator should consider whether large language model providers such as OpenAI and Anthropic should fall within its remit when their models play a significant role in co
Why it matters
If AI chatbots become the front door to banking and investing, the companies behind them may be regulated like financial firms, a shift worth watching for anyone heading into finance.
The three-week experiment in government-gated AI ended quietly this week: GPT-5.6 reached the public with an approval stamp rather than a standoff. But the story that opened as regulation closed as litigation, with Apple suing the partner it once built ChatGPT into iPhones with. Between the Geneva dialogue, the FCA's warning that chatbot makers may need regulating like financial firms, and a courtroom fight over whose secrets live in which researchers' heads, the industry's defining conflicts are now about jurisdiction: who gets to set the rules, and where.
JADEPUFFER gave all of that urgency a concrete face. The same agentic capability that GPT-5.6 sells as a feature ran a ransomware operation with no skilled human behind it. And the money kept pointing at physical limits: Europe's biggest fusion check, with Google on the cap table, is a bet that the real bottleneck for all of this is energy.
What to watch
Apple's case now heads to the Northern California federal court, where OpenAI's formal response will show whether it fights, settles, or countersues; the filings will also reveal how much of the io Products hardware program becomes public.
Test yourself
A few quick questions on this week. Just for you, nothing is recorded.
1. What cleared the way for GPT-5.6's public release on July 9?
2. What is the core allegation in Apple's lawsuit against OpenAI?
3. What made the JADEPUFFER ransomware attack a first, according to Sysdig?
4. Why is Proxima Fusion's 411 million euro round notable beyond its size?