OpenAI Limits GPT-5.6 Sol to Trusted Partners; Chip Stocks Hit Steepest Selloff in Months

US government access controls on frontier AI tighten while semiconductor stocks log their worst single-week decline in months.

🏛️ OpenAI Limits GPT-5.6 Sol, Terra, and Luna to Trusted Partners at Government Request

Decoded: OpenAI on June 26 released three new AI models — GPT-5.6 Sol, Terra, and Luna — and simultaneously complied with a US government request to limit initial access to a "small group of trusted partners." Sol is OpenAI's most capable model to date, with advances in coding, biology, and cybersecurity. OpenAI confirmed Sol does not cross its "critical" cybersecurity risk threshold and is more effective at fixing vulnerabilities than executing end-to-end attacks. The company previewed the models with the Trump administration before launch and described the access restriction as a "short-term step" toward full availability in coming weeks. The action follows Anthropic's June 12 suspension of Fable 5 and Mythos 5 access under a separate export control directive. (CNBC, OpenAI blog post, June 26, 2026)

Why it matters: The Trump administration is institutionalizing pre-release government access review as a condition for frontier model launches — a framework now applying to both Anthropic and OpenAI. OpenAI's stated goal of building a "repeatable process for future model releases" signals this is becoming standard, not exceptional. For enterprise developers, the effect is delayed access to the most capable models until government review clears. For investors, the more critical signal is that OpenAI is actively shaping the regulatory framework alongside the administration, increasing the probability that US frontier AI regulation takes a form the industry can manage rather than one driven by Congress or foreign bodies.

📊 Global Chip Stocks Post Worst Selloff in Months as AI Infrastructure Cost Concerns Bite

Decoded: Semiconductor stocks fell broadly on June 26 on investor concerns that rising AI infrastructure costs are squeezing technology sector margins. Intel fell 3%, Sandisk dropped 10%, Arm lost 4%, Marvell declined 5%, and Micron dropped more than 5% despite reporting blockbuster third-quarter earnings the prior day. Asian chip stocks fell harder: SK Hynix shed 8%, Samsung Electronics dropped 5%, and Japan's Advantest declined 10%. In Europe, ASML fell 2%, Infineon and ASM International each dropped 4%, and STMicroelectronics shed 4%. Separately, Oracle posted its worst weekly stock performance since the 2001 dot-com bust, falling 19% on concerns over its $130 billion debt load — accumulated primarily to fund AI data center construction for OpenAI — and negative free cash flow of nearly $24 billion in its latest fiscal year. (CNBC, June 26, 2026)

Why it matters: The AI infrastructure cost reality is arriving simultaneously across the semiconductor supply chain and the data center build-out. Chip stocks fell on the same day Micron reported record earnings driven by AI memory demand — indicating investors are not doubting demand, but questioning whether AI infrastructure margins can sustain the capital intensity required to meet it. Oracle's 19% weekly collapse and $130 billion debt position illustrate the structural risk: companies that committed heavily to AI infrastructure before revenue fully materialized now face severe balance sheet pressure. For investors, the simultaneous selloff across semiconductors (INTC, ARM, MU), Asian chip producers (SK Hynix, Samsung), European suppliers (ASML), and AI infrastructure operators (ORCL) signals a market repricing of AI infrastructure as lower-return-on-capital than previously assumed — a meaningful shift in the sector's risk/reward profile.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team