OpenAI Launches GPT-5.6 Under Unprecedented Government Restrictions
In a move that sent shockwaves through the AI industry, OpenAI released its GPT-5.6 model family on Friday — but only to a select group of approved partners, complying with a direct request from the Trump administration to limit initial access.
The lineup consists of three models: Sol, the flagship with enhanced agentic capabilities in coding, biology, and cybersecurity; Terra, a balanced everyday model performing on par with GPT-5.5 at half the cost; and Luna, a fast, affordable option for high-volume workloads.
Sol introduces a new "ultra" reasoning mode that coordinates multiple subagents for complex tasks, and OpenAI claims it outperforms Anthropic's Claude Mythos 5 on coding workflows while using one-third fewer output tokens. Pricing ranges from $1/$6 per million tokens (Luna) up to $5/$30 (Sol).
The restricted rollout is the first time the U.S. government has directly intervened in a major AI model launch. OpenAI pushed back in its announcement, stating: "We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
The company says it expects to make the models broadly available "within weeks" through ChatGPT, Codex, and the API, as it works with the administration on a longer-term cybersecurity framework.
Sources: TechCrunch, CNBC
OpenAI and Broadcom Unveil Jalapeño, a Custom AI Inference Chip
Just days before the GPT-5.6 announcement, OpenAI made another major move: revealing Jalapeño, its first custom-designed silicon for AI inference, built in partnership with Broadcom.
The chip went from initial design to manufacturing tape-out in just nine months — what the companies claim is the fastest ASIC development cycle ever achieved in high-performance semiconductors. Notably, OpenAI's own models were used to accelerate parts of the chip design and optimization process.
Broadcom CEO Hock Tan said early testing shows roughly 50% cost savings compared to current state-of-the-art GPUs, with substantially better performance per watt. The chip is designed specifically for the inference workload of serving LLMs to users, rather than the training process.
Initial deployment is planned for late 2026. The move signals OpenAI's ambition to reduce its dependence on Nvidia's GPU ecosystem and control more of its infrastructure stack — a critical strategic play as the company serves over 900 million weekly active users through ChatGPT.
Sources: TechCrunch, VentureBeat
AI Stocks in Freefall: Wall Street Questions the $452 Billion AI Bet
The Nasdaq Composite posted its fifth consecutive losing session on Friday as investors continued to dump AI and tech stocks, deepening a sell-off that has wiped more than 6% from the index since its all-time high on June 2.
The numbers are stark: Alphabet plunged 6-10% over two sessions, Amazon dropped 4%, and chipmaker Micron tumbled 13% before partially recovering. The trigger? Growing anxiety that the AI industry's massive capital expenditures aren't translating into proportional revenue.
Combined 2026 capex across Microsoft, Alphabet, Amazon, and Meta has exceeded $452 billion, a figure that institutional investors are increasingly calling unsustainable without clearer paths to AI monetization. Reports suggest OpenAI is even considering delaying its planned IPO to next year due to market volatility and SpaceX's disappointing post-IPO performance.
The sell-off has rippled globally, with South Korea's KOSPI index plunging 10% in a single session earlier in the week, driven largely by declines in AI-exposed semiconductor stocks.
Sources: CNN Business, Intellectia, NPR
Industry Giants Unite Behind Agentic Resource Discovery Standard
While the market panics, the AI industry is quietly building the infrastructure for the next phase. On June 17, Google and Microsoft introduced the Agentic Resource Discovery (ARD) specification — an open standard that lets AI agents find, verify, and connect to tools, APIs, and other agents across the web.
Released under the Apache 2.0 license, ARD defines two key primitives: a static ai-catalog.json manifest hosted at a well-known path on an organization's domain, and a registry API that indexes these catalogs and returns ranked matches to natural-language discovery queries.
The specification addresses a critical missing piece in the AI agent ecosystem: how does an agent discover what capabilities exist and verify they're safe to use? Until now, agents have relied on hardcoded tool lists or manual configuration.
Launch contributors include Cisco, Databricks, GitHub, GoDaddy, Google, Hugging Face, Microsoft, Nvidia, Salesforce, ServiceNow, and Snowflake — a notably broad coalition that spans cloud, developer tools, and enterprise software.
Sources: Google Developers Blog, Microsoft Command Line, Hugging Face
Colorado Rewrites Its AI Law Days Before It Was Set to Take Effect
In a last-minute regulatory twist, Colorado's landmark AI Act will not take effect on its scheduled date of June 30. Governor Jared Polis signed SB 189 on May 14, which repeals and replaces the original 2024 law and pushes the effective date to January 1, 2027.
The original Colorado AI Act was the first comprehensive state-level AI regulation in the U.S., imposing duties of care, risk management programs, and impact assessments on companies using AI in consequential decisions about employment, housing, healthcare, and education.
The replacement law significantly scales back these requirements, shifting from a duty-of-care framework to a disclosure-based approach with limited individual rights in narrow circumstances. The move reflects intense lobbying pressure from the tech industry and growing concerns among legislators that overly prescriptive regulation could drive AI companies out of the state.
Meanwhile, at the federal level, bipartisan legislation — the Great American Artificial Intelligence Act of 2026 — is working its way through Congress, aiming to create the first comprehensive national framework for AI governance.
Sources: Hunton Andrews Kurth, Troutman Pepper