2026-07-08
By Vadym · Generated with AI, curated by me
Everyone’s quietly rewriting their own rules this week — a safety watchdog says the major labs are moving the goalposts on commitments they made themselves, Congress is trying to slow a flood of cheap Chinese models it can’t actually un-release, and Microsoft cut nearly 5,000 jobs without once saying the word “automation.” Musk, meanwhile, just merged an AI company into a rocket company and called it a rebrand.
The House Committee on Homeland Security and the Select Committee on the CCP, led by Reps. Moolenaar and Garbarino, are jointly investigating the national-security risk of Chinese-developed models — including DeepSeek, Alibaba’s Qwen, and Moonshot AI’s Kimi — after sending letters to Cursor and Airbnb over their use. The probe follows data showing US companies’ share of tokens routed to Chinese models has stayed above 30% every week since February, spiking as high as 46%, up from just 11% the year before, as those models run 60–90% cheaper for comparable work.
The Future of Life Institute’s latest AI Safety Index graded Anthropic, OpenAI, Google DeepMind, and Meta across 37 indicators in six categories, using seven outside reviewers plus public disclosures. None scored above a C+, and the panel said all four are “moving the goalposts” — quietly weakening commitments they’d made in earlier assessment cycles. FLI chair Max Tegmark said the industry is “sprinting toward a cliff” while continuing to race toward more powerful systems despite acknowledging the risk.
GitHub shipped the Copilot desktop app to every plan, including the free tier and GitHub Education, letting anyone run agent-driven coding sessions on macOS, Windows, or Linux — even without a subscription, by bringing your own model key. The same week, Moonshot AI’s open-weight Kimi K2.7 expanded from Copilot’s Pro tiers to Business and Enterprise, becoming the first open-weight model selectable across GitHub’s entire lineup, while Gemini 2.5 Pro and Gemini 3 Flash are set to be dropped from Copilot entirely on July 31.
Shenzhen-based Even Realities raised $150 million in a pre-Series B led by Meituan and Tencent, valuing the three-year-old startup at $1 billion. Founded by ex-Apple engineers, the company built a proprietary waveguide display system called Even HAO that beams information straight into the wearer’s line of sight — deliberately skipping the camera that rivals like Meta’s Ray-Bans are built around.
Microsoft eliminated roughly 4,800 roles, about 2.1% of its global workforce, in cuts the company says weren’t a direct AI swap but reflect how “AI is changing how work gets done” by automating routine tasks. The cut adds to a running 2026 tally of roughly 120,000 tech layoffs where employers have cited AI as a factor — even as many of those same companies report record revenue.
Elon Musk’s xAI officially rebranded as SpaceXAI, completing the all-stock absorption into SpaceX first announced in February and folding Grok, the X platform, and xAI’s infrastructure roadmap under the rocket company’s umbrella with a new logo. The move lands the same week SpaceX’s stock, ticker SPCX, joins the Nasdaq-100, following a June IPO that valued the combined company at $1.77 trillion.
Today’s throughline is who actually holds the leverage. Congress can investigate Chinese models but can’t un-open-source them. Four labs can grade their own safety and still fail the grade they wrote. Microsoft can automate work without ever using that word. Even the boldest move today — Musk folding an AI company into a rocket company — is a bet on how markets categorize things, not a bet on new capability. Nobody in this brief is fully in control of the story their own AI investment is telling.
— Boba
Curated by Vadym