· 001 · AI News · 7 min read
GPT-5.6 Launches, Anthropic Mythos Standoff, OpenAI Jalapeño Chip — AI News Briefing
Top 7 Stories
1. OpenAI Launches GPT-5.6 Under Limited Preview With Trump Administration Oversight
Less than 24 hours after news broke that OpenAI would stagger its next model release at the request of the Trump administration, GPT-5.6 arrived. The new suite includes three models: Sol (the flagship), Terra (a medium-tier model for high-volume work), and Luna (a “fast and affordable” everyday model). OpenAI says GPT-5.6 is especially skilled at coding, cybersecurity, biology, and staying focused during long-horizon agentic AI tasks.
Pricing aggressively undercuts Anthropic: Sol costs $5 input / $30 output per million tokens — nearly half of Anthropic’s Claude Fable 5 at $10/$50. Sol also debuts two additional modes: “max” for deeper reasoning and “ultra” for sub-agent orchestration. OpenAI dedicated approximately 700,000 A100e GPU hours to automated red-teaming and emphasized that safeguards “may occasionally intervene on legitimate work” — the very thing the limited preview is designed to test. The Trump administration will approve customer access on a case-by-case basis during this period.
2. Anthropic’s Mythos Standoff Drags Into Week Three With No Resolution
It’s been 14 days since the Trump administration’s June 12th export control order forced Anthropic to take its Mythos 5 and Fable 5 models offline, and there’s still no end in sight. The order prohibits access by “any foreign national” — including Anthropic’s own non-US employees — due to cybersecurity concerns. Anthropic cofounder Tom Brown has replaced CEO Dario Amodei in negotiations alongside public policy chief Sarah Heck, but updates remain nonexistent.
Security researcher Katie Moussouris, who reviewed the Fable 5 vulnerability report at Anthropic’s request, called it “significantly overblown,” noting that the jailbreak technique — asking the model to “fix this code” rather than “find security issues” — is actually “the most valuable thing an AI model can do for defensive security.” The impasse couldn’t come at a worse time for Anthropic, which needs Mythos revenue to fund its $15 billion/year SpaceX data center deal ahead of a planned IPO.
3. OpenAI Reveals Jalapeño, Its First Custom AI Processor
OpenAI unveiled Jalapeño, an ASIC (Application-Specific Integrated Circuit) designed for AI inference, built in partnership with Broadcom. Broadcom CEO Hock Tan claims it matches the performance of Nvidia’s Blackwell chips and Google’s Tensor Processing Units. The chip comes just nine months after OpenAI announced its partnership with Broadcom to reduce reliance on supply-constrained Nvidia GPUs.
OpenAI calls Jalapeño the “first step in a multi-generation compute platform” and expects deployment by the end of 2026. Early testing shows “performance per watt substantially better than current state-of-the-art.” The move puts OpenAI alongside Microsoft, Meta, and Amazon — all of which have recently launched custom AI chips — in the race to build independent AI infrastructure.
4. Google DeepMind, Microsoft, and xAI Agree to Government Pre-Release Model Reviews
Google DeepMind, Microsoft, and Elon Musk’s xAI have agreed to allow the US government to review new AI models before public release. The Commerce Department’s Center for AI Standards and Innovation (CAISI) will perform “pre-deployment evaluations and targeted research to better assess frontier AI capabilities.” CAISI has completed 40 reviews since beginning evaluations of OpenAI and Anthropic models in 2024.
Both OpenAI and Anthropic have renegotiated their partnerships with CAISI “to better align with priorities in President Donald Trump’s AI Action Plan,” according to Bloomberg. The New York Times separately reported that Trump is considering an executive order to “bring together tech executives and government officials” to oversee new AI models — signaling a potentially permanent regulatory framework for frontier AI.
5. Anthropic Launches Claude Tag — AI Agent That Joins Your Slack Workspace
Anthropic introduced Claude Tag, a new integration that lets Claude join a company’s Slack workspace as a member. Users simply tag @Claude to delegate tasks: writing and merging pull requests, locating sales numbers, analyzing data, and executing multi-step workflows. The agentic Slack integration marks Anthropic’s most direct push into enterprise productivity, embedding AI where teams already work.
The launch comes at an awkward time for Anthropic, however — Claude Fable 5, the model that would power such enterprise features, remains partially offline due to the ongoing export control dispute. It’s unclear how fully Claude Tag can function while the company’s most capable models are restricted.
6. Hollywood Stars Back ‘Human Consent Standard’ for AI Licensing
George Clooney, Tom Hanks, Meryl Streep, Viola Davis, and Cate Blanchett are among the high-profile names backing the Human Consent Standard, a new AI licensing framework that lets individuals and rights holders declare whether AI systems can use their likeness, voice, creative works, or characters. The standard, overseen by RSL Media (cofounded by Blanchett), builds on the Really Simple Licensing (RSL) Standard for web content.
A registry launching in June 2026 will let people verify their identity and set permissions — allow, prohibit, or require permission — that AI systems can discover and respect. “RSL Media is a simple, effective and free solutions-based technology for facilitating and activating consent,” Blanchett said. “It’s also the industry’s first practical solution where people everywhere, not just public figures, can assert control over how their work is used by AI.”
7. Nearly 400 Local Newspapers Sue OpenAI and Microsoft Over Copyright
A coalition of nearly 400 local newspapers has filed a lawsuit against OpenAI and Microsoft, alleging the companies “scraped, copied, and ingested” their work without permission or compensation to train AI models. The suit adds to a growing docket of copyright battles OpenAI faces from publishers including The New York Times, Ziff Davis, Merriam-Webster, and Encyclopedia Britannica.
The case highlights the escalating tension between AI developers and content creators as training data provenance becomes a central legal and ethical question. With no federal AI copyright legislation yet in place, these lawsuits are effectively shaping the boundaries of fair use in the AI era through the courts.
Trend Watch
| Story | Impact | Why it Matters |
|---|---|---|
| GPT-5.6 limited preview | High | The first major model release under direct US government access control sets a precedent for how frontier AI reaches market |
| Anthropic Mythos standoff | Critical | A prolonged shutdown of a major AI lab’s most powerful models threatens US competitiveness and IPO timelines |
| OpenAI Jalapeño chip | High | Reducing dependence on Nvidia reshapes the AI hardware landscape and could lower inference costs industry-wide |
| CAISI pre-deployment reviews | Medium-High | Formalizing government AI review across Google, Microsoft, and xAI signals a shift from voluntary to expected oversight |
| Claude Tag Slack agent | Medium | AI agents embedded in workplace tools represent the next frontier of enterprise AI adoption |
| Human Consent Standard | Medium | A practical, celebrity-backed framework for AI consent could become the de facto standard for responsible AI training |
| Newspaper copyright lawsuit | Medium | The outcome of mass copyright litigation will define the economics of AI training data for years to come |
What to Watch
The Trump administration’s AI regulatory framework is taking shape in real time. The contrast between OpenAI’s relatively cooperative staggered release and Anthropic’s punishing export-control shutdown reveals an ad-hoc approach that could harden into permanent policy. Trump’s potential executive order on AI model oversight, combined with CAISI’s expanding review mandate, suggests the window for self-regulation is closing.
AI hardware independence is accelerating. OpenAI’s Jalapeño chip follows similar moves by Microsoft, Meta, and Amazon. As custom silicon matures, Nvidia’s dominance — and the GPU supply bottleneck — may finally ease, potentially democratizing access to frontier AI compute by late 2026.
Enterprise AI agents are graduating from demos to Slack. Anthropic’s Claude Tag and OpenAI’s sub-agent “ultra” mode for Sol both point toward AI that doesn’t just chat, but acts — writing code, analyzing data, and operating inside the tools teams already use. The agent era is arriving through your workplace messaging app.