THE RUNDOWN

The AI lab's annualized revenue figure, previously reported at around $70 billion, is now being revised sharply downward to roughly $50 billion, according to a new report. That's still a staggering number by any normal measure, but in a company that's raised hundreds of billions in investment at sky-high valuations, a $20 billion miss matters. It raises real questions about whether the AI boom's commercial returns are keeping pace with its ambitions.

Three OpenAI safety researchers who were let go have written an open letter disputing the company's misconduct claims against them. More importantly, they're warning that their dismissals are having a chilling effect on internal safety culture — researchers are reportedly becoming reluctant to raise concerns. Given how central trust is to the AI industry's long-term credibility, this is a dispute worth watching closely.

Google is pushing Gemini well beyond a chatbot. The updated system can now plan and execute multi-step tasks, delegate work to sub-agents, and operate across business apps and data systems autonomously. It's Google's clearest signal yet that the future of enterprise AI isn't about answering questions — it's about doing the work. Businesses are first in line, but consumer rollouts will likely follow.

Anthropic has launched OSS Scanner, a free tool that helps open-source projects find security vulnerabilities using AI. Open-source software powers most of the internet's infrastructure, and it's chronically under-resourced when it comes to security review. Offering this at no cost is a genuinely useful move that could have outsized impact on software safety across the board.

SpaceX has quietly acquired a portfolio of low-band spectrum licenses and is now publicly declaring its intention to build Starlink Mobile into a serious mobile carrier. Low-band spectrum travels further and penetrates buildings better than higher frequencies — it's the backbone of reliable nationwide coverage. For markets across Africa where terrestrial network gaps are real, a competitive satellite-based mobile option from SpaceX could be genuinely significant.

In a separate but related development, SpaceX has called for better coordination between satellite operators after a series of near-misses in orbit involving its own Starlink constellation. The company described conjunctions — close approaches between objects — of 'tens of meters to hundreds of meters,' calling it 'way too close to comfort.' As the number of satellites in low-Earth orbit multiplies, collision avoidance is becoming one of the more urgent and underreported tech challenges of this decade.

Bloomberg is reporting that Apple will unveil a new MacBook Pro with a touchscreen display, along with an updated iPad Mini, on or around October 27th. It's a feature Windows laptops have had for years, and Apple users have been asking about for nearly as long. Whether Apple's version is a polished improvement on the concept or just a belated catch-up will depend entirely on the software experience it ships with.

IDC's latest data shows worldwide PC shipments fell 20 percent in Q3 compared to the same period last year, with analysts warning that worsening macro conditions could make the next few quarters harder still. Consumer spending on hardware is clearly under pressure globally, which makes the context for that touchscreen MacBook launch more interesting — premium devices need a buyer who isn't tightening their belt.

In a policy update that's partly practical and partly philosophical, Anthropic has added a rule prohibiting 'sustained and needless abusive or cruel behavior' directed at its AI models. The company also tightened its election-related policies ahead of the midterms. It's a reminder that as AI systems become more sophisticated, the companies building them are having to make active choices about what kinds of interactions they'll permit — and what that says about how we should think about these systems at all. Check out our explainer on AI ethics and policy if you want the bigger picture.

The company behind LMArena — the popular crowdsourced leaderboard where users vote on which AI model gives better answers — has raised $200 million from Lightspeed and Khosla Ventures, pushing its valuation to $3.1 billion in under a year. The platform is now expanding into measuring AI models on alignment and safety, not just raw performance. In a world increasingly skeptical of AI benchmarks, a human-preference-based ranking system clearly has serious commercial appeal.

It's a genuinely complex day in tech — big money, safety concerns, and some very real infrastructure bets all competing for attention at once. The OpenAI revenue story and the fired researchers' letter together make for an uncomfortable picture of an industry that's still figuring out whether its promises are matching its reality.